Search

Using programmable system-on-chip to implement computer vision functions poses many challenges due to highly constrained resources in cost, size and power consumption. In this work, we propose a new neuro-inspired image processing model and implemented it on a system-on-chip Xilinx Z702c board. With the attractor neural network model to store the object’s contour information, we eliminate the computationally expensive steps in the curve evolution re-initialisation at every new iteration or frame. Our experimental results demonstrate that this integrated approach achieves accurate and robust object tracking, when they are partially or completely occluded in the scenes. Importantly, the system is able to process 640 by 480 videos in real-time stream with 30 frames per second using only one low-power Xilinx Zynq-7000 system-on-chip platform. This proof-of-concept work has demonstrated the advantage of incorporating neuro-inspired features in solving image processing problems during occlusion. [ABSTRACT FROM AUTHOR]

In high-resolution remote sensing image processing, segmentation is a crucial step that extracts information within the object-based image analysis framework. Because of its robustness, mean-shift segmentation algorithms are widely used in the field of image segmentation. However, the traditional implementation of these methods cannot process large volumes of images rapidly under limited computing resources. Currently, parallel computing models are generally employed for segmentation tasks with massive remote sensing images. This paper presents a parallel implementation of the mean-shift segmentation algorithm based on an analysis of the principle and characteristics of this technique. To avoid the inconsistency on the boundaries of adjacent data chunks, we propose a novel buffer-zone-based data-partitioning strategy. Employing the proposed data-partitioning strategy, two intensively computation steps are performed in parallel on different data chunks. The experimental results show that the proposed algorithm effectively improves the computing efficiency of image segmentation in a parallel computing environment. Furthermore, they demonstrate the practicality of massive image segmentation when computer resources are limited. [ABSTRACT FROM AUTHOR]

Remote photoplethysmography (rPPG) can achieve contactless human vital signs monitoring, but its signal quality is limited by the remote operation nature. In practical applications, improving the rPPG signal quality becomes an essential task. As a remote imaging technique, rPPG utilizes a camera to capture a video of a skin area, especially the facial area, then focuses on a particular sub-area as the region of interest (ROI). In this paper, we investigated a novel adaptive ROI (AROI) approach for improving the rPPG signal quality. In this approach, block-based spatial-temporal division is performed on a captured face video. Based on these segmented video pipelines, the spatial-temporal quality distribution of the rPPG signals is estimated using a signal-to-noise ratio (SNR) feature. Afterwards, AROIs are calculated through mean-shift clustering and adaptive thresholding in SNR maps. As the AROI can be dynamically adjusted according to the spatial-temporal quality distribution of rPPG signals on the face, the quality of the final recovered rPPG signal is improved. The performance of the proposed AROI approach was evaluated with both still and moving subjects. Compared to conventional ROI methods for rPPG, the proposed AROI obtained a higher accuracy in heart rate measurement. And the state-of-the-art motion-resistant rPPG techniques can be effectively enhanced through being integrated with the AROI. [ABSTRACT FROM AUTHOR]

Image segmentation is essential to image processing because it provides a solution to the task of separating the objects in an image from the background and from each other, which is an important step in object recognition, tracking, and other high-level image-processing applications. By partitioning the input image into smaller regions, segmentation performs the balancing act of extracting the main areas of interest (objects and important features) that further help to interpret the image, while remaining immune to irrelevant noise and less important background scenes. Image-segmentation applications branch off into a plethora of domains, from decision-making applications in computer vision to medical imaging and quality control to name just a few. The mean-shift algorithm provides a unique unsupervised clustering solution to image segmentation, and it has an established record of good performance for a wide variety of input images. However, mean-shift segmentation exhibits an unfavorable computational complexity of O(kN2), where N represents the number of pixels and k the number of iterations. As a result of this complexity, unsupervised image segmentation has had limited impact in autonomous applications, where a low-power, real-time solution is required. We propose a novel hardware architecture that exploits the customizable computing power of FPGAs and reduces the execution time by clustering pixels in parallel while meeting the low-power demands of embedded applications. The architecture performance is compared with existing CPU and GPU implementations to demonstrate its advantages in terms of both execution time and energy. [ABSTRACT FROM AUTHOR]

Unique and stable reference point is essential for registration and identification in automated fingerprint identification systems. Most existing methods for detecting reference points need to scan the fingerprint image or orientation field pixel by pixel or block by block to confirm a candidate reference point. The inherent complexity of this process makes those methods time-consuming. In this paper, we propose a two-step method to improve the efficiency of detecting reference points by (1) determining the singular point, i.e., the approximate position of the reference point, in a novel fast way; then (2) refining the reference point precisely in the local area of the singular point. In the first step, a walking algorithm is proposed which can walk directly to the singular point without scanning the whole fingerprint image and hence it is extremely fast. Then, in the local area around the singular point, an enhanced method based on mean-shift concept (EMS-based method) is designed to localize the reference point precisely. Experimental results on FVC2000 DB1a and DB2a databases validate that the proposed WEMS (Walking + EMS) method outperforms two state-of-the-art methods in terms of accuracy and efficiency. [ABSTRACT FROM AUTHOR]

Spatiogram is a generalization of the histogram. It adds high-order spatial information so that the target can be described more precisely. It is important to choose a suitable method of measuring the similarity between two spatiograms when Spatiogram is applied to the target tracking field. However, the original similarity measure based on spatiogram has the limitation of the insufficient discriminative power. Therefore, in this paper, we propose a new spatiogram similarity measure method called BJSD (Bhattacharyya coefficient and Jensen-Shannon divergence). The similarity of the color feature and the similarity of the spatial distribution are calculated by the Bhattacharyya coefficient (BC) and Jensen-Shannon divergence (JSD). Experiments show that the improved similarity measure has better discriminative than the Conaire’s method and the tracking results are more stable and accurate than the traditional mean shift tracking method. [ABSTRACT FROM AUTHOR]

Object tracking is a critical task in surveillance and activity analysis. Two main issues for tracking are appearance (illumination) and structural (size of a target) variations of the object. We propose a method which is robust and addresses these issues by incorporating features that are less variant to these changes. The proposed features are mean local binary pattern (mLBP), an illumination invariant texture feature, and modified fuzzy c-means (MFCM) weighted color histogram to handle both illumination and scale changes. These features are combined to form a hybrid mean-shift (MS) vector and used in the MS vector framework for target tracking. Experimental results using standard benchmark videos show that the proposed scheme can lead to better localization and robust tracking in challenging illumination scenarios, when compared to several existing tracking algorithms. [ABSTRACT FROM AUTHOR]

Effective appearance models are one critical factor for robust object tracking. In this paper, we introduce foreground feature saliency concept into the background modelling, and put forward a novel foreground saliency-based background-weighted histogram scheme (FSBWH) for target representation and tracking, which exploits salient features from both foreground and background. We think that background and foreground salient features are both crucial for target representation and tracking. Experimental results show that the proposed FSBWH scheme can improve the robustness and performance of tracker significantly especially in complex occlusions and similar background scenes. [ABSTRACT FROM AUTHOR]

The mean-shift algorithm is a popular algorithm in computer vision and image processing. It can also be cast as a minimum gamma-divergence estimation. In this paper we focus on the 'blurring' mean-shift algorithm, which is one version of the mean-shift process that successively blurs the dataset. The analysis of the blurring mean-shift is relatively more complicated compared to the nonblurring version, yet the algorithm convergence and the estimation consistency have not been well studied in the literature. In this paper we prove both the convergence and the consistency of the blurring mean-shift. We also perform simulation studies to compare the efficiency of the blurring and the nonblurring versions of the mean-shift algorithms. Our results show that the blurring mean-shift has more efficiency. [ABSTRACT FROM AUTHOR]

In this paper, the fuzzy coding histogram representation for an image region is proposed for visual tracking. At the initial frame, we use the fuzzy clustering technology on all pixels within the initial detection box to get the cluster prototypes, which are used to construct the fuzzy codebook. During the tracking period, the candidate image region is also represented by a fuzzy coding histogram which is constructed by the fuzzy memberships. This representation can incorporate more feature information and enhance the tracking robustness. In addition, we use the cumulative distribution function between the fuzzy coding histograms to construct a cross-bin metric and use the mean-shift iteration to realize the robust visual tracking. Finally, some tracking experimental results are presented to demonstrate the effectiveness of the proposed approach. [ABSTRACT FROM AUTHOR]

Object tracking is a fundamental challenge in computer vision. For real-time tracking, the efficiency and robustness of the Mean-shift algorithm makes it a popular choice. However, the scale of the Mean-shift kernel is a crucial parameter and no clear mechanism exists presently for updating the scale when a size-changing object is tracked. In this paper, a new method is introduced using projective geometry to determine the size of the object, and in turn the scale of the Mean-shift kernel. In the initial step of the algorithm, the geometric information of the scene is obtained automatically (or manually). With the geometric information, the size of the object is updated. The experimental results show that this algorithm is stable, efficient and outperforms the Mean-shift baseline and other kernel updating methods, such as CAMSHIFT. [ABSTRACT FROM AUTHOR]

Recently, particle filter has been applied to many visual tracking problems and it has been modified in order to reduce the computation time or memory usage. One of them is the Mean-Shift embedded particle filter (MSEPF, for short) and it is further modified as Randomized MSEPF. These methods can decrease the number of the particles without the loss of tracking accuracy. However, the accuracy may depend on the definition of the likelihood function (observation model) and of the prediction model. In this paper, the authors propose an extension of these models in order to increase the tracking accuracy. Furthermore, the expansion resetting method, which was proposed for mobile robot localization, and the changing the size of the window in Mean-Shift search are also selectively applied in order to treat the occlusion or rapid change of the movement. [ABSTRACT FROM AUTHOR]

Deformable model fitting has been actively pursued in the computer vision community for over a decade. As a result, numerous approaches have been proposed with varying degrees of success. A class of approaches that has shown substantial promise is one that makes independent predictions regarding locations of the model's landmarks, which are combined by enforcing a prior over their joint motion. A common theme in innovations to this approach is the replacement of the distribution of probable landmark locations, obtained from each local detector, with simpler parametric forms. In this work, a principled optimization strategy is proposed where nonparametric representations of these likelihoods are maximized within a hierarchy of smoothed estimates. The resulting update equations are reminiscent of mean-shift over the landmarks but with regularization imposed through a global prior over their joint motion. Extensions to handle partial occlusions and reduce computational complexity are also presented. Through numerical experiments, this approach is shown to outperform some common existing methods on the task of generic face fitting. [ABSTRACT FROM AUTHOR]

Existing water hazard detection methods usually fail when the features of water surfaces are greatly changed by the surroundings, e.g., by a change in illumination. This paper proposes a novel algorithm to robustly detect different kinds of water hazards for autonomous navigation. Our algorithm combines traditional machine learning and image segmentation and uses only digital cameras, which are usually affordable, as the visual sensors. Active learning is used for automatically dealing with problems caused by the selection, labeling and classification of large numbers of training sets. Mean-shift based image segmentation is used to refine the final classification. Our experimental results show that our new algorithm can accurately detect not only ‘common’ water hazards, which usually have the features of both high brightness and low texture, but also ‘special’ water hazards that may have lots of ripples or low brightness. [ABSTRACT FROM AUTHOR]

The classic mean-shift tracker has no integrated scale adaptation, which limits its performance in tracking variable scale object as wel l as the object with severe motions. Based on the variation analysis of Bhattacharyya coefficient within mean-shift framework, the sufficient conditions for accurate tracking of object with scale changes are presented. We propose that the changes of object scale and position within the region of previous tracking window will not impact the localization accuracy of mean-shift tracker. Based on our findings, a novel backward tracking method is introduced to solve scaling problem, and the solution of dealing with the severe object motions is also discussed by integrating mean-shift tracker into the low-resolution matching scheme. [ABSTRACT FROM AUTHOR]