Search

Many image-based rendering (IBR) methods rely on depth estimates obtained from structured light or time-of-flight depth sensors to synthesize novel views from sparse camera networks. However, these estimates often contain missing or noisy regions, resulting in an incorrect mapping between source and target views. This situation makes the fusion process more challenging, as the visual information is misaligned, inconsistent, or missing. In this work, we first implement a lightweight network based on the transformer, which is well-known for its capability to model long-range relationships within input data, to extract spatial features from color images. These features are then used to enhance the quality of completed depth maps. Furthermore, we combine a sequential deep neural network with a spatial attention mechanism to effectively fuse the projected features from multiple source viewpoints. This approach enables us to integrate information from an arbitrary number of source viewpoints as well as improve accuracy in synthesized views. Experimental results on challenging datasets demonstrate that our method achieves superior synthesized image quality compared to state-of-the-art (SOTA) methods.

Recent advancements in high dynamic range (HDR) display technology have significantly enhanced the contrast ratios and peak brightness of modern displays. In the coming years, it is expected that HDR televisions capable of delivering significantly higher brightness and, therefore, contrast levels than today’s models will become increasingly accessible and affordable to consumers. While HDR technology has gained prominence over the past few years, low dynamic range (LDR) content is still consumed due to a substantial volume of historical multimedia content being recorded and preserved in LDR. Although the amount of HDR content will continue to increase as HDR becomes more prevalent, a large portion of multimedia content currently remains in LDR. In addition, it is worth noting that although the HDR standard supports multimedia content with luminance levels up to 10,000 cd/m2 (a standard measure of brightness), most HDR content is typically limited to a maximum brightness of around 1000 cd/m2. This limitation aligns with the current capabilities of consumer HDR TVs but is a factor approximately five times brighter than current LDR TVs. To accurately present LDR content on a HDR display, it is processed through a dynamic range expansion process known as inverse tone mapping (iTM). This LDR to HDR conversion faces many challenges, including the inducement of noise artifacts, false contours, loss of details, desaturated colors, and temporal inconsistencies. This paper introduces complete inverse tone mapping, artifact suppression, and a highlight enhancement pipeline for video sequences designed to address these challenges. Our LDR-to-HDR technique is capable of adapting to the peak brightness of different displays, creating HDR video sequences with a peak luminance of up to 6000 cd/m2. Furthermore, this paper presents the results of comprehensive objective and subjective experiments to evaluate the effectiveness of the proposed pipeline, focusing on two primary aspects: real-time operation capability and the quality of the HDR video output. Our findings indicate that our pipeline enables real-time processing of Full HD (FHD) video (1920 × 1080 pixels), even on hardware that has not been optimized for this task. Furthermore, we found that when applied to existing HDR content, typically capped at a brightness of 1000 cd/m2, our pipeline notably enhances its perceived quality when displayed on a screen that can reach higher peak luminances.

Hyperspectral image (HSI) processing tasks frequently rely on spatial–spectral total variation (SSTV) to quantify the local smoothness of image structures. However, conventional SSTV only considers a sparse structure of gradient maps computed along the spatial and spectral dimensions while neglecting other correlations. To address this limitation, we introduce low-rank guided SSTV (LRSTV), which characterizes the sparsity and low-rank priors of the gradient map simultaneously. First, we verify through numerical tests and theoretical analyses that the gradient tensors are not only sparse but also low-rank. Subsequently, to model the low rankness of the gradient map, we use the tensor average rank to represent the low Tucker rank of gradient tensors. The convex envelope of the tensor average rank is then employed to penalize the rank on the gradient map after the Fourier transform along the spectral dimension. By naturally encoding the sparsity and low-rank priors of the gradient map, LRSTV results in a more accurate representation of the original image. Finally, we demonstrate the effectiveness of LRSTV by integrating it into the HSI processing model, replacing conventional SSTV, and testing it on two public datasets with nine cases of mixed noise and two datasets with realistic noise. The results show that LRSTV outperforms conventional SSTV in terms of accuracy and robustness.

Extracting information of interest from continuous video streams is a strongly demanded computer vision task. For the realization of this task at the edge using the current de-facto standard approach, i.e., deep learning, it is critical to optimize key performance metrics such as throughput and energy consumption according to prescribed application requirements. This allows achieving timely decision-making while extending the battery lifetime as much as possible. In this context, we propose a method to boost neural-network performance based on a co-execution strategy that exploits hardware heterogeneity on edge platforms. The enabling tool is Apache TVM, a highly efficient machine-learning compiler compatible with a diversity of hardware back-ends. The proposed approach solves the problem of network partitioning and distributes the workloads to make concurrent use of all the processors available on the board following a pipeline scheme. We conducted experiments on various popular CNNs compiled with TVM on the Jetson TX2 platform. The experimental results based on measurements show a significant improvement in throughput with respect to a single-processor execution, ranging from 14% to 150% over all tested networks. Power-efficient configurations were also identified, accomplishing energy reductions above 10%.

In this paper, we propose a new cooperative method that improves the accuracy of Turn Movement Count (TMC) under challenging conditions by introducing contextual observations from the surrounding areas. The proposed method focuses on the correct identification of the movements in conditions where current methods have difficulties. Existing vision-based TMC systems are limited under heavy traffic conditions. The main problems for most existing methods are occlusions between vehicles that prevent the correct detection and tracking of the vehicles through the entire intersection and the assessment of the vehicle’s entry and exit points, incorrectly assigning the movement. The proposed method intends to overcome this incapability by sharing information with other observation systems located at neighboring intersections. Shared information is used in a cooperative scheme to infer the missing data, thereby improving the assessment that would otherwise not be counted or miscounted. Experimental evaluation of the system shows a clear improvement over related reference methods.

The manufacturing of metal parts via powder-bed fusion is often still facing quality issues due to microstructural porosity. Minimizing this porosity remains a priority and requires the optimization of printing process parameters. While the analysis of printed parts using X-ray computed tomography can localize and identify the pore types (e.g. keyhole or lack-of-fusion pores), these pore types can be difficult to identify across printer settings and print materials. Therefore, there is a need for a material and process agnostic approach. This work presents such an approach by considering a set of geometric pore features that do not differ considerably across print scenarios. These features are then leveraged for supervised pore type classification. The distributions of pore features were analyzed in different materials and under varying laser parameters, showing that they behave in a generic way. For classification, it is observed that they outperform other features leveraged in the state-of-the-art for pore classification in a single material, reaching up to 93.0% accuracy. Additionally, accuracies up to 90.2% for cross-material classification were observed by training on pores of one material and validating on another. These results pave the way to a general-purpose pore classification method usable across materials and process conditions.

High dynamic range (HDR) imaging technology is increasingly being used in automated driving systems (ADS) for improving the safety of traffic participants in scenes with strong differences in illumination. Therefore, a combination of HDR video, that is video with details in all illumination regimes, and (HDR) object perception techniques that can deal with this variety in illumination is highly desirable. Although progress has been made in both HDR imaging solutions and object detection algorithms in the recent years, they have progressed independently of each other. This has led to a situation in which object detection algorithms are typically designed and constantly improved to operate on 8 bit per channel content. This makes these algorithms not ideally suited for use in HDR data processing, which natively encodes to a higher bit-depth (12 bits/16 bits per channel). In this paper, we present and evaluate two novel convolutional neural network (CNN) architectures that intelligently convert high bit depth HDR images into 8-bit images. We attempt to optimize reconstruction quality by focusing on ADS object detection quality. The first research novelty is to jointly perform tone-mapping with demosaicing by additionally successfully suppressing noise and demosaicing artifacts. The first CNN performs tone-mapping with noise suppression on a full-color HDR input, while the second performs joint demosaicing and tone-mapping with noise suppression on a raw HDR input. The focus is to increase the detectability of traffic-related objects in the reconstructed 8-bit content, while ensuring that the realism of the standard dynamic range (SDR) content in diverse conditions is preserved. The second research novelty is that for the first time, to the best of our knowledge, a thorough comparative analysis against the state-of-the-art tone-mapping and demosaicing methods is performed with respect to ADS object detection accuracy on traffic-related content that abounds with diverse challenging (i.e., boundary cases) scenes. The evaluation results show that the two proposed networks have better performance in object detection accuracy and image quality, than both SDR content and content obtained with the state-of-the-art tone-mapping and demosaicing algorithms.

Limitation of labeled samples has always been a challenge for hyperspectral image (HSI) classification. In real remote sensing applications, we encounter a situation where an HSI scene is not labeled at all. To solve this problem, cross-domain learning methods are developed by utilizing another HSI scene with similar land covers and sufficient labeled samples. However, the disparity between HSI scenes is still a challenge in reducing the classification performance, which may be affected by variations in illumination and weather. As a robust supplement to these variations, light detection and ranging (LiDAR) data provide stable elevation and spatial information. In this article, we propose a multisource cross-domain classification method using fractional fusion and spatial-spectral domain adaptation to reduce the disparity between scenes. First, the spatial information of HSI is preserved by fractional differential masks. Then, the LiDAR data are utilized for spectral alignment of HSI. The utilization of LiDAR data reduces the pixel-level disparity between scenes. At last, a spatial-spectral domain adaptation network is proposed to reduce domain shift at the feature level and extract discriminative spatial-spectral features. Experimental results on HSI and LiDAR scenes show 5% –10% improvements in overall accuracy compared with the state-of-the-art methods.

Intelligent driver assistance systems are becoming increasingly popular in modern passenger vehicles. A crucial component of intelligent vehicles is the ability to detect vulnerable road users (VRUs) for an early and safe response. However, standard imaging sensors perform poorly in conditions of strong illumination contrast, such as approaching a tunnel or at night, due to their dynamic range limitations. In this paper, we focus on the use of high-dynamic-range (HDR) imaging sensors in vehicle perception systems and the subsequent need for tone mapping of the acquired data into a standard 8-bit representation. To our knowledge, no previous studies have evaluated the impact of tone mapping on object detection performance. We investigate the potential for optimizing HDR tone mapping to achieve a natural image appearance while facilitating object detection of state-of-the-art detectors designed for standard dynamic range (SDR) images. Our proposed approach relies on a lightweight convolutional neural network (CNN) that tone maps HDR video frames into a standard 8-bit representation. We introduce a novel training approach called detection-informed tone mapping (DI-TM) and evaluate its performance with respect to its effectiveness and robustness in various scene conditions, as well as its performance relative to an existing state-of-the-art tone mapping method. The results show that the proposed DI-TM method achieves the best results in terms of detection performance metrics in challenging dynamic range conditions, while both methods perform well in typical, non-challenging conditions. In challenging conditions, our method improves the detection F2 score by 13%. Compared to SDR images, the increase in F2 score is 49%.

Introduction: Accurately detecting cracks is crucial for assessing the health of materials. Manual detection methods are time-consuming, leading to the development of automatic detection techniques based on image processing and machine learning. These methods utilize morphological image processing and material deformation analysis through Digital Image or Volume Correlation techniques (DIC/DVC) to identify cracks. The strain field derived from DIC/DVC tends to be noisy. Traditional denoising methods sacrifice spatial resolution, limiting their effectiveness in capturing abrupt structural deformations such as fractures. Method: In this study, a novel DVC regularization method is proposed to obtain a sharper and less noisy strain field. The method minimizes the total variation of spatial strain field components based on the assumption of approximate strain constancy within material phases. Results: The proposed methodology is validated using simulated data and actual 4D μ-CT experimental data. Compared to classical denoising methods, the proposed DVC regularization method provides a more reliable crack detection with fewer false positives. Conclusions: These results highlight the possibility of estimating a low-noise strain field without relying on the spatial smoothness assumption, thereby improving accuracy and reliability in crack detection.

Digital volume correlation (DVC) is a 3D image-based technique for displacement and strain computation. Traditionally, both (digital image correlation) DIC and DVC are methods based on two individual time frames; the estimation of the displacement and strain field is done using one reference and one moving frame as input. However, dynamic experiments generate more than two temporal frames. Therefore, with classical DVC techniques, only a subset of the available data is used. In this study, we propose a novel DVC method that can rely on more than two frames for the displacement and strain computation. The proposed method aims to be as general as possible; there is no constraint regarding the nature or the rate of the displacement (e.g., cyclic or linear). The aim of this method is to impose a temporal regularization that improves the self-consistency of the algorithm. The multi-frame DVC improves the quality of the registration in challenging situations. As an example, we investigate the dissolution of a pharmaceutical tablet in water, which undergoes three processes: swelling, gel formation, and material erosion. The accuracy of the registration—quantified by the sum of square differences (SSD)—has improved by 23% on an average with respect to the classical two-frame method. Classical DVC methods fail in registering images with structures that change appearance through time, such as the tablet that, in contact with water, reacts chemically, changing phase and becoming a gel. Moreover, we proved that multi-frame DVC is more robust in registering images with severe but realistic motion artefacts. As an example for this case, we apply the method to a series of μ-CT datasets of aluminum foam during a compression experiment. As seen with the tablets, we are in a situation where the appearance of the structures in the images changes through time, but in this case it is because of motion artefacts. Finally, the use of more than two frames makes the method more robust against noisy images, with an average improvement of 35% in registration accuracy obtained using the three-frame DVC method compared to the classical two-frame DVC method.

This paper reveals the characteristics and effects of nonverbal behavior and human mimicry in the context of application interviews. It discloses a novel analyzation method for psychological research by utilizing machine learning. In comparison to traditional manual data analysis, machine learning proves to be able to analyze the data more deeply and to discover connections in the data invisible to the human eye. The paper describes an experiment to measure and analyze the reactions of evaluators to job applicants who adopt specific behaviors: mimicry, suppress, immediacy and natural behavior. First, evaluation of the applicant qualifications by the interviewer reveals how behavioral self-management can improve the interviewer’s opinion of the candidate. Secondly, the underlying mechanics of mimicry behavior are exposed through analysis of seven nonverbal actions. Manual data analysis determines the frequency features of the actions and answers how often the actions are performed and how often they are mimicked during application interviews. Two of the seven actions are here deemed negligible due too low frequency features. Finally, machine learning is employed to analyze the data in great detail and distinguish the four behavior categories from each other. A Random Forest classifier is able to achieve 55.2% accuracy for predicting the behavior condition of the interviews while human observers reach an accuracy of 32.9%. The feature set for the classifier is reduced to 130 features with the most important features relating to the correlations between the leaning forward actions of the interview participants.

Over the last two decades, UAVs have become an indispensable acquisition platform in the remote sensing community. Meanwhile, advanced lightweight sensors have been introduced in the market, including LiDAR scanners with multiple beams and hyperspectral cameras measuring reflectance using many different narrow-banded filters. To date, however, few fully fledged drone systems exist that combine different sensing modalities in a way that complements the strengths and weaknesses of each. In this paper, we present our multimodal drone payload and sensor fusion pipeline, which allows multispectral point clouds to be generated at subcentimeter accuracy. To that end, we combine high-frequency navigation outputs from a professional-grade GNSS with photogrammetric bundle adjustment and a dedicated point cloud registration algorithm that takes full advantage of LiDAR’s specifications. We demonstrate that the latter significantly improves the quality of the reconstructed point cloud in terms of fewer ghosting effects and less noise. Finally, we thoroughly discuss the impact of the quality of the GNSS/INS system on the structure from the motion and LiDAR SLAM reconstruction process.

We propose a novel and efficient algorithm for detection of specular reflections and light sources (highlights) in cinematic content. The detection of highlights is important for reconstructing them properly in the conversion of the low dynamic range (LDR) to high dynamic range (HDR) content. Highlights are often difficult to be distinguished from bright diffuse surfaces, due to their brightness being reduced in the conventional LDR content production. Moreover, the cinematic LDR content is subject to the artistic use of effects that change the apparent brightness of certain image regions (e.g. limiting depth of field, grading, complex multi-lighting setup, etc.). To ensure the robustness of highlights detection to these effects, the proposed algorithm goes beyond considering only absolute brightness and considers five different features. These features are: the size of the highlight relative to the size of the surrounding image structures, the relative contrast in the surrounding of the highlight, its absolute brightness expressed through the luminance (luma feature), through the saturation in the color space (maxRGB feature) and through the saturation in white (minRGB feature). We evaluate the algorithm on two different image data-sets. The first one is a publicly available LDR image data-set without cinematic content, which allows comparison to the broader State of the art. Additionally, for the evaluation on cinematic content, we create an image data-set consisted of manually annotated cinematic frames and real-world images. For the purpose of demonstrating the proposed highlights detection algorithm in a complete LDR-to-HDR conversion pipeline, we additionally propose a simple inverse-tone-mapping algorithm. The experimental analysis shows that the proposed approach outperforms conventional highlights detection algorithms on both image data-sets, achieves high quality reconstruction of the HDR content and is suited for use in LDR-to-HDR conversion.

Large 3D electron microscopy data sets frequently contain noisy data due to accelerated imaging, and denoising techniques require specialised skill sets. Here the authors introduce DenoisEM, an ImageJ plugin that democratises denoising EM data sets, enabling fast parameter tuning and processing through parallel computing.

Pedestrian detection is an important research domain due to its relevance for autonomous and assisted driving, as well as its applications in security and industrial automation. Often, more than one type of sensor is used to cover a broader range of operating conditions than a single-sensor system would allow. However, it remains difficult to make pedestrian detection systems perform well in highly dynamic environments, often requiring extensive retraining of the algorithms for specific conditions to reach satisfactory accuracy, which, in turn, requires large, annotated datasets captured in these conditions. In this paper, we propose a probabilistic decision-level sensor fusion method based on naive Bayes to improve the efficiency of the system by combining the output of available pedestrian detectors for colour and thermal images without retraining. The results in this paper, obtained through long-term experiments, demonstrate the efficacy of our technique, its ability to work with non-registered images, and its adaptability to cope with situations when one of the sensors fails. The results also show that our proposed technique improves the overall accuracy of the system and could be very useful in several applications.

Multi-exposure image fusion (MEF) methods for high dynamic range (HDR) imaging suffer from ghosting artifacts when dealing with moving objects in dynamic scenes. The state-of-the-art methods use optical flow to align low dynamic range (LDR) images before merging, introducing distortion into the aligned LDR images from inaccurate motion estimation due to large motion and occlusion. In place of pre-alignment, attention-based methods calculate the correlation between the reference LDR image and non-reference LDR images, thus excluding misaligned regions in LDR images. Nevertheless, they also exclude the saturated details at the same time. Taking advantage of both the alignment and attention-based methods, we propose an efficient Deep HDR Deghosting Fusion Network (DDFNet) guided by optical flow and image correlation attentions. Specifically, the DDFNet estimates the optical flow of the LDR images by a motion estimation module and encodes that optical flow as a flow feature. Additionally, it extracts correlation features between the reference LDR and other non-reference LDR images. The optical flow and correlation features are employed to adaptably combine information from LDR inputs in an attention-based fusion module. Following the merging of features, a decoder composed of Dense Networks reconstructs the HDR image without ghosting. Experimental results indicate that the proposed DDFNet achieves state-of-the-art image fusion performance on different public datasets.

Dual cameras with visible-thermal multispectral pairs provide both visual and thermal appearance, thereby enabling detecting pedestrians around the clock in various conditions and applications, including autonomous driving and intelligent transportation systems. However, due to the greatly varying real-world scenarios, the performance of a detector trained on a source dataset might change dramatically when evaluated on another dataset. A large amount of training data is often necessary to guarantee the detection performance in a new scenario. Typically, human annotators need to conduct the data labeling work, which is time-consuming, labor-intensive and unscalable. To overcome the problem, we propose a novel unsupervised transfer learning framework for multispectral pedestrian detection, which adapts a multispectral pedestrian detector to the target domain based on pseudo training labels. In particular, auxiliary detectors are utilized and different label fusion strategies are introduced according to the estimated environmental illumination level. Intermediate domain images are generated by translating the source images to mimic the target ones, acting as a better starting point for the parameter update of the pedestrian detector. The experimental results on the KAIST and FLIR ADAS datasets demonstrate that the proposed method achieves new state-of-the-art performance without any manual training annotations on the target data.

Urban air quality mapping has been widely applied in urban planning, air pollution control and personal air pollution exposure assessment. Urban air quality maps are traditionally derived using measurements from fixed monitoring stations. Due to high cost, these stations are generally sparsely deployed in a few representative locations, leading to a highly generalized air quality map. In addition, urban air quality varies rapidly over short distances (2 by 100 m and 1 h resolution for fine-grained air quality mapping. The CLADF model exploits deep forest to construct a local model for each cluster consisting of nearest neighbor measurements in contextual feature space, and considers traffic flow as an important contextual feature. Extensive validation experiments were conducted using mobile NO2 measurements collected by 17 postal vans equipped with low-cost sensors operating in Antwerp, Belgium. The experimental results demonstrate that the CLADF model achieves the lowest RMSE as well as advances in accuracy and correlation, compared with various benchmark models, including random forest, deep forest, extreme gradient boosting and support vector regression.

In laser powder bed fusion (LPBF), melt pool instability can lead to the development of pores in printed parts, reducing the part’s structural strength. While camera-based monitoring systems have been introduced to improve melt pool stability, these systems only measure melt pool stability in limited, indirect ways. We propose that melt pool stability can be improved by explicitly encoding stability into LPBF monitoring systems through the use of temporal features and pore density modelling. We introduce the temporal features, in the form of temporal variances of common LPBF monitoring features (e.g., melt pool area, intensity), to explicitly quantify printing stability. Furthermore, we introduce a neural network model trained to link these video features directly to pore densities estimated from the CT scans of previously printed parts. This model aims to reduce the number of online printer interventions to only those that are required to avoid porosity. These contributions are then implemented in a full LPBF monitoring system and tested on prints using 316L stainless steel. Results showed that our explicit stability quantification improved the correlation between our predicted pore densities and true pore densities by up to 42%.

In this paper, we propose a unified and flexible framework for general image fusion tasks, including multi-exposure image fusion, multi-focus image fusion, infrared/visible image fusion, and multi-modality medical image fusion. Unlike other deep learning-based image fusion methods applied to a fixed number of input sources (normally two inputs), the proposed framework can simultaneously handle an arbitrary number of inputs. Specifically, we use the symmetrical function (e.g., Max-pooling) to extract the most significant features from all the input images, which are then fused with the respective features from each input source. This symmetry function enables permutation-invariance of the network, which means the network can successfully extract and fuse the saliency features of each image without needing to remember the input order of the inputs. The property of permutation-invariance also brings convenience for the network during inference with unfixed inputs. To handle multiple image fusion tasks with one unified framework, we adopt continual learning based on Elastic Weight Consolidation (EWC) for different fusion tasks. Subjective and objective experiments on several public datasets demonstrate that the proposed method outperforms state-of-the-art methods on multiple image fusion tasks.

Today, solar energy is taking an increasing share of the total energy mix. Unfortunately, many operational photovoltaic plants suffer from a plenitude of defects resulting in non-negligible power loss. The latter highly impacts the overall performance of the PV site; therefore, operators need to regularly inspect their solar parks for anomalies in order to prevent severe performance drops. As this operation is naturally labor-intensive and costly, we present in this paper a novel system for improved PV diagnostics using drone-based imagery. Our solution consists of three main steps. The first step locates the solar panels within the image. The second step detects the anomalies within the solar panels. The final step identifies the root cause of the anomaly. In this paper, we mainly focus on the second step comprising the detection of anomalies within solar panels, which is done using a region-based convolutional neural network (CNN). Experiments on six different PV sites with different specifications and a variety of defects demonstrate that our anomaly detector achieves a true positive rate or recall of more than 90% for a false positive rate of around 2% to 3% tested on a dataset containing nearly 9000 solar panels. Compared to the best state-of-the-art methods, the experiments revealed that we achieve a slightly higher true positive rate for a substantially lower false positive rate, while tested on a more realistic dataset.

In magnetic resonance (MR) images, detection of focal cortical dysplasia (FCD) lesion as a main pathological cue of epilepsy is challenging because of the variability in the presentation of FCD lesions. Existing algorithms appear to have sufficient sensitivity in detecting lesions but also generate large numbers of false-positive (FP) results. In this paper, we propose a multiple classifier fusion and optimization schemes to automatically detect FCD lesions in MR images with reduced FPs through constructing an objective function based on the F-score. Thus, the proposed scheme obtains an improved tradeoff between minimizing FPs and maximizing true positives. The optimization is achieved by incorporating the genetic algorithm into the work scheme. Hence, the contribution of weighting coefficients to different classifications can be effectively determined. The resultant optimized weightings are applied to fuse the classification results. A set of six typical FCD features and six corresponding Z-score maps are evaluated through the mean F-score from multiple classifiers for each feature. From the experimental results, the proposed scheme can automatically detect FCD lesions in 9 out of 10 patients while correctly classifying 31 healthy controls. The proposed scheme acquires a lower FP rate and a higher F-score in comparison with two state-of-the-art methods.

Close-range hyperspectral images are a promising source of information in plant biology, in particular, for in vivo study of physiological changes. In this study, we investigate how data fusion can improve the detection of leaf elements by combining pixel reflectance and morphological information. The detection of image regions associated to the leaf structures is the first step toward quantitative analysis on the physical effects that genetic manipulation, disease infections, and environmental conditions have in plants. We tested our fusion approach on Musa acuminata (banana) leaf images and compared its discriminant capability to similar techniques used in remote sensing. Experimental results demonstrate the efficiency of our fusion approach, with significant improvements over some conventional methods.

Depth sensing has improved rapidly in recent years, which allows for structural information to be utilized in various applications, such as virtual reality, scene and object recognition, view synthesis, and 3D reconstruction. Due to the limitations of the current generation of depth sensors, the resolution of depth maps is often still much lower than the resolution of color images. This hinders applications, such as view synthesis or 3D reconstruction, from providing high-quality results. Therefore, super-resolution, which allows for the upscaling of depth maps while still retaining sharpness, has recently drawn much attention in the deep learning community. However, state-of-the-art deep learning methods are typically designed and trained to handle a fixed set of integer-scale factors. Moreover, the raw depth map collected by the depth sensor usually has many depth data missing or misestimated values along the edges and corners of observed objects. In this work, we propose a novel deep learning network for both depth completion and depth super-resolution with arbitrary scale factors. The experimental results on the Middlebury stereo, NYUv2, and Matterport3D datasets demonstrate that the proposed method can outperform state-of-the-art methods.

This paper proposes a method to infer road networks from GPS traces. These networks include intersections between roads, the connectivity between the intersections and the possible traffic directions between directly-connected intersections. These intersections are localized by detecting and clustering turning points, which are locations where the moving direction changes on GPS traces. We infer the structure of road networks by segmenting all of the GPS traces to identify these intersections. We can then form both a connectivity matrix of the intersections and a small representative GPS track for each road segment. The road segment between each pair of directly-connected intersections is represented using a series of geographical locations, which are averaged from all of the tracks on this road segment by aligning them using the dynamic time warping (DTW) algorithm. Our contribution is two-fold. First, we detect potential intersections by clustering the turning points on the GPS traces. Second, we infer the geometry of the road segments between intersections by aligning GPS tracks point by point using a “stretch and then compress” strategy based on the DTW algorithm. This approach not only allows road estimation by averaging the aligned tracks, but also a deeper statistical analysis based on the individual track’s time alignment, for example the variance of speed along a road segment.

This paper presents a vulnerable road user (VRU) tracking algorithm capable of handling noisy and missing detections from heterogeneous sensors. We propose a cooperative fusion algorithm for matching and reinforcing of radar and camera detections using their proximity and positional uncertainty. The belief in the existence and position of objects is then maximized by temporal integration of fused detections by a multi-object tracker. By switching between observation models, the tracker adapts to the detection noise characteristics making it robust to individual sensor failures. The main novelty of this paper is an improved imputation sampling function for updating the state when detections are missing. The proposed function uses a likelihood without association that is conditioned on the sensor information instead of the sensor model. The benefits of the proposed solution are two-fold: firstly, particle updates become computationally tractable and secondly, the problem of imputing samples from a state which is predicted without an associated detection is bypassed. Experimental evaluation shows a significant improvement in both detection and tracking performance over multiple control algorithms. In low light situations, the cooperative fusion outperforms intermediate fusion by as much as 30%, while increases in tracking performance are most significant in complex traffic scenes.

The simultaneous acquisition of multi-spectral images on a single sensor can be efficiently performed by single shot capture using a mutli-spectral filter array. This paper focused on the demosaicing of color and near-infrared bands and relied on a convolutional neural network (CNN). To train the deep learning model robustly and accurately, it is necessary to provide enough training data, with sufficient variability. We focused on the design of an efficient training procedure by discovering an optimal training dataset. We propose two data selection strategies, motivated by slightly different concepts. The general term that will be used for the proposed models trained using data selection is data selection-based multi-spectral demosaicing (DSMD). The first idea is clustering-based data selection (DSMD-C), with the goal to discover a representative subset with a high variance so as to train a robust model. The second is an adaptive-based data selection (DSMD-A), a self-guided approach that selects new data based on the current model accuracy. We performed a controlled experimental evaluation of the proposed training strategies and the results show that a careful selection of data does benefit the speed and accuracy of training. We are still able to achieve high reconstruction accuracy with a lightweight model.

With the rapid development in sensing technology, data mining, and machine learning fields for human health monitoring, it became possible to enable monitoring of personal motion and vital signs in a manner that minimizes the disruption of an individual’s daily routine and assist individuals with difficulties to live independently at home. A primary difficulty that researchers confront is acquiring an adequate amount of labeled data for model training and validation purposes. Therefore, activity discovery handles the problem that activity labels are not available using approaches based on sequence mining and clustering. In this paper, we introduce an unsupervised method for discovering activities from a network of motion detectors in a smart home setting. First, we present an intra-day clustering algorithm to find frequent sequential patterns within a day. As a second step, we present an inter-day clustering algorithm to find the common frequent patterns between days. Furthermore, we refine the patterns to have more compressed and defined cluster characterizations. Finally, we track the occurrences of various regular routines to monitor the functional health in an individual’s patterns and lifestyle. We evaluate our methods on two public data sets captured in real-life settings from two apartments during seven-month and three-month periods.

Drafting involves cycling so close behind another person that wind resistance is significantly reduced, which is illegal during most long distance and several short distance triathlon and duathlon events. In this paper, a proof of concept for a drafting detection system based on computer vision is proposed. After detecting and tracking a bicycle through the various scenes, the distance to this object is estimated through computational geometry. The probability of drafting is then determined through statistical analysis of subsequent measurements over an extended period of time. These algorithms are tested using a static recording and a recording that simulates a race situation with ground truth distances obtained from a Light Detection And Ranging (LiDAR) system. The most accurate developed distance estimation method yields an average error of 0.46 m in our test scenario. When sampling the distances at periods of 1 or 2 s, simulations demonstrate that a drafting violation is detected quickly for cyclists riding at 2 m or more below the limit, while generally avoiding false positives during the race-like test set-up and five hour race simulations.

Extrinsic camera calibration is essential for any computer vision task in a camera network. Typically, researchers place a calibration object in the scene to calibrate all the cameras in a camera network. However, when installing cameras in the field, this approach can be costly and impractical, especially when recalibration is needed. This paper proposes a novel, accurate and fully automatic extrinsic calibration framework for camera networks with partially overlapping views. The proposed method considers the pedestrians in the observed scene as the calibration objects and analyzes the pedestrian tracks to obtain extrinsic parameters. Compared to the state of the art, the new method is fully automatic and robust in various environments. Our method detect human poses in the camera images and then models walking persons as vertical sticks. We apply a brute-force method to determines the correspondence between persons in multiple camera images. This information along with 3D estimated locations of the top and the bottom of the pedestrians are then used to compute the extrinsic calibration matrices. We also propose a novel method to calibrate the camera network by only using the top and centerline of the person when the bottom of the person is not available in heavily occluded scenes. We verified the robustness of the method in different camera setups and for both single and multiple walking people. The results show that the triangulation error of a few centimeters can be obtained. Typically, it requires less than one minute of observing the walking people to reach this accuracy in controlled environments. It also just takes a few minutes to collect enough data for the calibration in uncontrolled environments. Our proposed method can perform well in various situations such as multi-person, occlusions, or even at real intersections on the street.

The Intelligent Transportation Systems area has experienced great developments in the recent past, although suffering from slow adoption ratios thus depriving consumers of many interesting and innovative applications. The only solution to this problem is to develop Intelligent Transportation Systems solutions using the already available technologies that are within the grasp of the common people, to make them cost-effective, quick to deploy and easy to adopt. We have therefore developed an affordable Intelligent Transportation Systems that make use of standard smartphones to assist drivers when overtaking. The system autonomously creates a network among the close-by vehicles and provides drivers with a real-time video feed from the one located just ahead. Our system seamlessly offers a better view of the road, and of any vehicle travelling in the opposite direction, being especially useful when the front view of the driver is blocked by large vehicles. We have validated our overtaking assistance system, in both laboratory environment and realistic scenarios. The laboratory tests involved choosing the most effective video codec between MJPEG and H.264, for providing real-time video streaming. Then, using the chosen codec, we performed the outdoor tests to further tune our application to maximise performance. The preliminary results from our experiments allow being optimistic about the effectiveness and applicability of the proposed system.

The blurred gray/white matter junction is an important feature of focal cortical dysplasia (FCD) lesions. FCD is the main cause of epilepsy and can be detected through magnetic resonance (MR) imaging. Several earlier studies have focused on computing the gradient magnitude of the MR image and used the resulting map to model the blurred gray/white matter junction. However, gradient magnitude cannot quantify the blurred gray/white matter junction. Therefore, we proposed a novel algorithm called local directional probability optimization (LDPO) for detecting and quantifying the width of the gray/white matter boundary (GWB) within the lesional areas. The proposed LDPO method mainly consists of the following three stages: (1) introduction of a hidden Markov random field-expectation-maximization algorithm to compute the probability images of brain tissues in order to obtain the GWB region; (2) generation of local directions from gray matter (GM) to white matter (WM) passing through the GWB, considering the GWB to be an electric potential field; (3) determination of the optimal local directions for any given voxel of GWB, based on iterative searching of the neighborhood. This was then used to measure the width of the GWB. The proposed LDPO method was tested on real MR images of patients with FCD lesions. The results indicated that the LDPO method could quantify the GWB width. On the GWB width map, the width of the blurred GWB in the lesional region was observed to be greater than that in the non-lesional regions. The proposed GWB width map produced higher F-scores in terms of detecting the blurred GWB within the FCD lesional region as compared to that of FCD feature maps, indicating better trade-off between precision and recall.

In this paper, we propose a novel extrinsic calibration method for camera networks using a sphere as the calibration object. First of all, we propose an easy and accurate method to estimate the 3D positions of the sphere center w.r.t. the local camera coordinate system. Then, we propose to use orthogonal procrustes analysis to pairwise estimate the initial camera relative extrinsic parameters based on the aforementioned estimation of 3D positions. Finally, an optimization routine is applied to jointly refine the extrinsic parameters for all cameras. Compared to existing sphere-based 3D position estimators which need to trace and analyse the outline of the sphere projection in the image, the proposed method requires only very simple image processing: estimating the area and the center of mass of the sphere projection. Our results demonstrate that we can get a more accurate estimate of the extrinsic parameters compared to other sphere-based methods. While existing state-of-the-art calibration methods use point like features and epipolar geometry, the proposed method uses the sphere-based 3D position estimate. This results in simpler computations and a more flexible and accurate calibration method. Experimental results show that the proposed approach is accurate, robust, flexible and easy to use.

This paper reports the outcomes of the 2014 Data Fusion Contest organized by the Image Analysis and Data Fusion Technical Committee (IADF TC) of the IEEE Geoscience and Remote Sensing Society (IEEE GRSS). As for previous years, the IADF TC organized a data fusion contest aiming at fostering new ideas and solutions for multisource remote sensing studies. In the 2014 edition, participants considered multiresolution and multisensor fusion between optical data acquired at 20-cm resolution and long-wave (thermal) infrared hyperspectral data at 1-m resolution. The Contest was proposed as a double-track competition: one aiming at accurate landcover classification and the other seeking innovation in the fusion of thermal hyperspectral and color data. In this paper, the results obtained by the winners of both tracks are presented and discussed.

This paper proposes an automated system for monitoring mobility patterns using a network of very low resolution visual sensors (30 × 30 pixels). The use of very low resolution sensors reduces privacy concern, cost, computation requirement and power consumption. The core of our proposed system is a robust people tracker that uses low resolution videos provided by the visual sensor network. The distributed processing architecture of our tracking system allows all image processing tasks to be done on the digital signal controller in each visual sensor. In this paper, we experimentally show that reliable tracking of people is possible using very low resolution imagery. We also compare the performance of our tracker against a state-of-the-art tracking method and show that our method outperforms. Moreover, the mobility statistics of tracks such as total distance traveled and average speed derived from trajectories are compared with those derived from ground truth given by Ultra-Wide Band sensors. The results of this comparison show that the trajectories from our system are accurate enough to obtain useful mobility statistics.

A vast array of pathologies is typified by the presence of nuclei with an abnormal morphology. Dysmorphic nuclear phenotypes feature dramatic size changes or foldings, but also entail much subtler deviations such as nuclear protrusions called blebs. Due to their unpredictable size, shape and intensity, dysmorphic nuclei are often not accurately detected in standard image analysis routines. To enable accurate detection of dysmorphic nuclei in confocal and widefield fluorescence microscopy images, we have developed an automated segmentation algorithm, called Blebbed Nuclei Detector (BleND), which relies on two-pass thresholding for initial nuclear contour detection, and an optimal path finding algorithm, based on dynamic programming, for refining these contours. Using a robust error metric, we show that our method matches manual segmentation in terms of precision and outperforms state-of-the-art nuclear segmentation methods. Its high performance allowed for building and integrating a robust classifier that recognizes dysmorphic nuclei with an accuracy above 95%. The combined segmentation-classification routine is bound to facilitate nucleus-based diagnostics and enable real-time recognition of dysmorphic nuclei in intelligent microscopy workflows.

The capture, processing and distribution of visual information is one of the major challenges for the paradigm of the Internet of Things. Privacy emerges as a fundamental barrier to overcome. The idea of networked image sensors pervasively collecting data generates social rejection in the face of sensitive information being tampered by hackers or misused by legitimate users. Power consumption also constitutes a crucial aspect. Images contain a massive amount of data to be processed under strict timing requirements, demanding high-performance vision systems. In this paper, we describe a hardware-based strategy to concurrently address these two key issues. By conveying processing capabilities to the focal plane in addition to sensing, we can implement privacy protection measures just at the point where sensitive data are generated. Furthermore, such measures can be tailored for efficiently reducing the computational load of subsequent processing stages. As a proof of concept, a full-custom QVGA vision sensor chip is presented. It incorporates a mixed-signal focal-plane sensing-processing array providing programmable pixelation of multiple image regions in parallel. In addition to this functionality, the sensor exploits reconfigurability to implement other processing primitives, namely block-wise dynamic range adaptation, integral image computation and multi-resolution filtering. The proposed circuitry is also suitable to build a granular space, becoming the raw material for subsequent feature extraction and recognition of categorized objects.

Hand tracking in video is an increasingly popular research field due to the rise of novel human-computer interaction methods. However, robust and real-time hand tracking in unconstrained environments remains a challenging task due to the high number of degrees of freedom and the non-rigid character of the human hand. In this paper, we propose an unsupervised method to automatically learn the context in which a hand is embedded. This context includes the arm and any other object that coherently moves along with the hand. We introduce two novel methods to incorporate this context information into a probabilistic tracking framework, and introduce a simple yet effective solution to estimate the position of the arm. Finally, we show that our method greatly increases robustness against occlusion and cluttered background, without degrading tracking performance if no contextual information is available. The proposed real-time algorithm is shown to outperform the current state-of-the-art by evaluating it on three publicly available video datasets. Furthermore, a novel dataset is created and made publicly available for the research community.

The 2013 Data Fusion Contest organized by the Data Fusion Technical Committee (DFTC) of the IEEE Geoscience and Remote Sensing Society aimed at investigating the synergistic use of hyperspectral and Light Detection And Ranging (LiDAR) data. The data sets distributed to the participants during the Contest, a hyperspectral imagery and the corresponding LiDAR-derived digital surface model (DSM), were acquired by the NSF-funded Center for Airborne Laser Mapping over the University of Houston campus and its neighboring area in the summer of 2012. This paper highlights the two awarded research contributions, which investigated different approaches for the fusion of hyperspectral and LiDAR data, including a combined unsupervised and supervised classification scheme, and a graph-based method for the fusion of spectral, spatial, and elevation information.

Reliable vision in challenging illumination conditions is one of the crucial requirements of future autonomous automotive systems. In the last decade, thermal cameras have become more easily accessible to a larger number of researchers. This has resulted in numerous studies which confirmed the benefits of the thermal cameras in limited visibility conditions. In this paper, we propose a learning-based method for visible and thermal image fusion that focuses on generating fused images with high visual similarity to regular truecolor (red-green-blue or RGB) images, while introducing new informative details in pedestrian regions. The goal is to create natural, intuitive images that would be more informative than a regular RGB camera to a human driver in challenging visibility conditions. The main novelty of this paper is the idea to rely on two types of objective functions for optimization: a similarity metric between the RGB input and the fused output to achieve natural image appearance; and an auxiliary pedestrian detection error to help defining relevant features of the human appearance and blending them into the output. We train a convolutional neural network using image samples from variable conditions (day and night) so that the network learns the appearance of humans in the different modalities and creates more robust results applicable in realistic situations. Our experiments show that the visibility of pedestrians is noticeably improved especially in dark regions and at night. Compared to existing methods we can better learn context and define fusion rules that focus on the pedestrian appearance, while that is not guaranteed with methods that focus on low-level image quality metrics.

Interpolation from a Color Filter Array (CFA) is the most common method for obtaining full color image data. Its success relies on the smart combination of a CFA and a demosaicing algorithm. Demosaicing on the one hand has been extensively studied. Algorithmic development in the past 20 years ranges from simple linear interpolation to modern neural-network-based (NN) approaches that encode the prior knowledge of millions of training images to fill in missing data in an inconspicious way. CFA design, on the other hand, is less well studied, although still recognized to strongly impact demosaicing performance. This is because demosaicing algorithms are typically limited to one particular CFA pattern, impeding straightforward CFA comparison. This is starting to change with newer classes of demosaicing that may be considered generic or CFA-agnostic. In this study, by comparing performance of two state-of-the-art generic algorithms, we evaluate the potential of modern CFA-demosaicing. We test the hypothesis that, with the increasing power of NN-based demosaicing, the influence of optimal CFA design on system performance decreases. This hypothesis is supported with the experimental results. Such a finding would herald the possibility of relaxing CFA requirements, providing more freedom in the CFA design choice and producing high-quality cameras.

In this paper, we describe a robust method for compensating the panning and tilting motion of a camera, applied to foreground−background segmentation. First, the necessary internal camera parameters are determined through feature-point extraction and tracking. From these parameters, two motion models for points in the image plane are established. The first model assumes a fixed tilt angle, whereas the second model allows simultaneous pan and tilt. At runtime, these models are used to compensate for the motion of the camera in the background model. We will show that these methods provide a robust compensation mechanism and improve the foreground masks of an otherwise state-of-the-art unsupervised foreground−background segmentation method. The resulting algorithm is always able to obtain F 1 scores above 80 % on every daytime video in our test set when a minimal number of only eight feature matches are used to determine the background compensation, whereas the standard approaches need significantly more feature matches to produce similar results.

Classification of hyperspectral images is a challenging task owing to the high dimensionality of the data, limited ground truth data, collinearity of the spectra and the presence of mixed pixels. Conventional classification techniques do not cope well with these problems. Thus, in addition to the spectral information, features were developed for a more complete description of the pixels, e.g., containing contextual information at the superpixel level or mixed pixel information at the subpixel level. This has encouraged an evolution of fusion techniques which use these myriad of multiple feature sets and decisions from individual classifiers to be employed in a joint manner. In this work, we present a flexible decision fusion framework addressing these issues. In a first step, we propose to use sparse fractional abundances as decision source, complementary to class probabilities obtained from a supervised classifier. This specific selection of complementary decision sources enables the description of a pixel in a more complete way, and is expected to mitigate the effects of small training samples sizes. Secondly, we propose to apply a fusion scheme, based on the probabilistic graphical Markov Random Field (MRF) and Conditional Random Field (CRF) models, which inherently employ spatial information into the fusion process. To strengthen the decision fusion process, consistency links across the different decision sources are incorporated to encourage agreement between their decisions. The proposed framework offers flexibility such that it can be extended with additional decision sources in a straightforward way. Experimental results conducted on two real hyperspectral images show superiority over several other approaches in terms of classification performance when very limited training data is available.

In this paper, we present a novel 2D–3D pedestrian tracker designed for applications in autonomous vehicles. The system operates on a tracking by detection principle and can track multiple pedestrians in complex urban traffic situations. By using a behavioral motion model and a non-parametric distribution as state model, we are able to accurately track unpredictable pedestrian motion in the presence of heavy occlusion. Tracking is performed independently, on the image and ground plane, in global, motion compensated coordinates. We employ Camera and LiDAR data fusion to solve the association problem where the optimal solution is found by matching 2D and 3D detections to tracks using a joint log-likelihood observation model. Each 2D–3D particle filter then updates their state from associated observations and a behavioral motion model. Each particle moves independently following the pedestrian motion parameters which we learned offline from an annotated training dataset. Temporal stability of the state variables is achieved by modeling each track as a Markov Decision Process with probabilistic state transition properties. A novel track management system then handles high level actions such as track creation, deletion and interaction. Using a probabilistic track score the track manager can cull false and ambiguous detections while updating tracks with detections from actual pedestrians. Our system is implemented on a GPU and exploits the massively parallelizable nature of particle filters. Due to the Markovian nature of our track representation, the system achieves real-time performance operating with a minimal memory footprint. Exhaustive and independent evaluation of our tracker was performed by the KITTI benchmark server, where it was tested against a wide variety of unknown pedestrian tracking situations. On this realistic benchmark, we outperform all published pedestrian trackers in a multitude of tracking metrics.

In this paper, we present a complete loop detection and correction system developed for data originating from lidar scanners. Regarding detection, we propose a combination of a global point cloud matcher with a novel registration algorithm to determine loop candidates in a highly effective way. The registration method can deal with point clouds that are largely deviating in orientation while improving the efficiency over existing techniques. In addition, we accelerated the computation of the global point cloud matcher by a factor of 2–4, exploiting the GPU to its maximum. Experiments demonstrated that our combined approach more reliably detects loops in lidar data compared to other point cloud matchers as it leads to better precision–recall trade-offs: for nearly 100% recall, we gain up to 7% in precision. Finally, we present a novel loop correction algorithm that leads to an improvement by a factor of 2 on the average and median pose error, while at the same time only requires a handful of seconds to complete.

Diffusion Weighted (DW) MRI allows for the non-invasive study of water diffusion inside living tissues. As such, it is useful for the investigation of human brain white matter (WM) connectivity in vivo through fiber tractography (FT) algorithms. Many DW-MRI tailored restoration techniques and FT algorithms have been developed. However, it is not clear how accurately these methods reproduce the WM bundle characteristics in real-world conditions, such as in the presence of noise, partial volume effect, and a limited spatial and angular resolution. The difficulty lies in the lack of a realistic brain phantom on the one hand, and a sufficiently accurate way of modeling the acquisition-related degradation on the other. This paper proposes a software phantom that approximates a human brain to a high degree of realism and that can incorporate complex brain-like structural features. We refer to it as a Diffusion BRAIN (D-BRAIN) phantom. Also, we propose an accurate model of a (DW) MRI acquisition protocol to allow for validation of methods in realistic conditions with data imperfections. The phantom model simulates anatomical and diffusion properties for multiple brain tissue components, and can serve as a ground-truth to evaluate FT algorithms, among others. The simulation of the acquisition process allows one to include noise, partial volume effects, and limited spatial and angular resolution in the images. In this way, the effect of image artifacts on, for instance, fiber tractography can be investigated with great detail. The proposed framework enables reliable and quantitative evaluation of DW-MR image processing and FT algorithms at the level of large-scale WM structures. The effect of noise levels and other data characteristics on cortico-cortical connectivity and tractography-based grey matter parcellation can be investigated as well.