Search

We present a new framework for online dense 3D reconstruction of indoor scenes by using only depth sequences. This research is particularly useful in cases with a poor light condition or in a nearly featureless indoor environment. The lack of RGB information makes long-range camera pose estimation difficult in a large indoor environment. The key idea of our research is to take advantage of the geometric prior of Manhattan scenes in each stage of the reconstruction pipeline with the specific aim to reduce the cumulative registration error and overall odometry drift in a long sequence. This idea is further boosted by local Manhattan frame growing and the local-to-global strategy that leads to implicit loop closure handling for a large indoor scene. Our proposed pipeline, namely ManhattanFusion, starts with planar alignment and local pose optimization where the Manhattan constraints are imposed to create detailed local segments. These segments preserve intrinsic scene geometry by minimizing the odometry drift even under complex and long trajectories. The final model is generated by integrating all local segments into a global volumetric representation under the constraint of Manhattan frame-based registration across segments. Our algorithm outperforms others that use depth data only in terms of both the mean distance error and the absolute trajectory error, and it is also very competitive compared with RGB-D based reconstruction algorithms. Moreover, our algorithm outperforms the state-of-the-art in terms of the surface area coverage by 10-40 percent, largely due to the usefulness and effectiveness of the Manhattan assumption through the reconstruction pipeline. more...

We propose a new method flow that utilizes pixel-level labeling information for instance-level object detection in indoor scenes from RGB-D data. Semantic labeling and instance segmentation are two different paradigms for indoor scene understanding that are usually accomplished separately and independently. We are interested in integrating the two tasks in a synergistic way in order to take advantage of their complementary nature for comprehensive understanding. Our work can capitalize on any deep learning networks used for semantic labeling by treating the intermediate layer as the category-wise local detection output, from which instance segmentation is optimized by jointly considering both the spatial fitness and the relational context encoded by three graphical models, namely, the vertical placement model (VPM), horizontal placement model (HPM) and non-placement model (NPM). VPM, HPM and NPM represent three common but distinct indoor object placement configurations: vertical, horizontal and hanging relationships, respectively. Experimental results on two standard RGB-D datasets show that our method can significantly improve small object segmentation with promising overall performance that is competitive with the state-of-the-art methods. more...

Objectives Furosemide is usually administered before the Coronary artery bypass grafting (CABG) to improve water–sodium retention. However, no final conclusions are available on the postoperative renal outcome of furosemide. We evaluated the effect of preoperative furosemide on acute kidney injury (AKI) after CABG. Methods We recorded the use of furosemide 14 days before surgery in all patients who underwent CABG from 2016 to 2017. Patients were divided into furosemide (F) group and non-furosemide (NF) group according to preoperative use of furosemide. A 1:1 propensity score matching was performed. Multivariate analyses were conducted to determine risk factors for AKI after CABG. Results Overall, 974 patients were included in the study, of which 82 cases were complicated with postoperative AKI. The incidence of AKI was significantly increased in F group than NF group (28.9% vs. 7.4%, p = 0.000). After adjusting for risk factors, the incidence of AKI in the F group was 5.34 times more than the NF group (95% confidence interval [CI] 2.45–11.64; p = 0.000). The incidence of AKI increased significantly when the cumulative dosage of furosemide exceeded 110 mg (odds ratio [OR] 6.23; 95% CI 2.07–18.74, p = 0.001) and 250 mg (OR 8.31; 95% CI 2.87–24.02, p = 0.000). After the propensity-matching group analysis, same results were obtained. Conclusions The incidence of AKI after CABG was related to the use of preoperative furosemide, and it increased exponentially with the increase of cumulative dose of furosemide. This provides guidance for the dose of preoperative furosemide. more...

Rare-class objects in natural scene images that are usually small and less frequent often convey more important information for scene understanding than the common ones. However, they are often overlooked in scene labeling studies due to two main reasons, low occurrence frequency and limited spatial coverage. Many methods have been proposed to enhance overall semantic labeling performance, but only a few consider rare-class objects. In this work, we present a deep semantic labeling framework with special consideration of rare classes via three techniques. First, a novel dual-resolution coarse-to-fine superpixel representation is developed, where fine and coarse superpixels are applied to rare classes and background areas respectively. This unique dual representation allows seamless incorporation of shape features into integrated global and local convolutional neural network (CNN) models. Second, shape information is directly involved during the CNN feature learning for both frequent and rare classes from the re-balanced training data, and also explicitly involved in data inference. Third, the proposed framework incorporates both shape information and the CNN architecture into semantic labeling through a fusion of probabilistic multi-class likelihood. Experimental results demonstrate competitive semantic labeling performance on two standard datasets both qualitatively and quantitatively, especially for rare-class objects. more...

Background Aberrant right subclavian artery (ARSA) with associated Kommerell diverticulum (KD) is a rare congenital aortic disease. KD patients have a high risk of rupture, dissection, and compression of adjacent structures. Although several treatment options have been proposed (traditional surgery, hybrid operation, and endovascular intervention), a consensus regarding optimal surgical management has not yet been established. Case presentation A case of successful hybrid repair of distal aortic arch dissection aneurysm by dissecting KD and ARSA with debranching of right and left common carotid arteries, left subclavian artery, and stent grafting was presented. Conclusions The hybrid operation is suitable for elderly patients or those with high risks. Along with intervention, the hybrid operation needs to be developed as a minimally invasive method. more...

Image patch priors become a popular tool for image denoising. The Gaussian mixture model (GMM) is remarkably effective in modelling natural image patches. However, GMM prior learning using the expectation maximisation (EM) algorithm is sensitive to the initialisation, often leading to low convergence rate of parameter estimation. In this study, a novel sampling method called random neighbourhood resampling (RNR) is proposed to improve the accuracy and efficiency of parameter estimation. An enhanced GMM (EGMM) learning algorithm is further developed by incorporating RNR into the EM algorithm to initialise and update the GMM prior. The learned EGMM prior is applied in the expected patch log-likelihood (EPLL) framework for image denoising. The effectiveness and performance of the proposed RNR and EGMM algorithm are demonstrated via extensive experimental results comparing with the state-of-the-art image denoising methods, the experimental results show the higher PSNR result of the denoised images using the proposed method. Meanwhile, the authors verified that the proposed method can efficiently reduce the time of image denoising compared with the basic EPLL method. more...

In the real world, many tasks require multiple agents to cooperate with each other under the condition of local observations. To solve such problems, many multi-agent reinforcement learning methods based on Centralized Training with Decentralized Execution have been proposed. One representative class of work is value decomposition, which decomposes the global joint Q-value $Q_\text{jt}$ into individual Q-values $Q_a$ to guide individuals' behaviors, e.g. VDN (Value-Decomposition Networks) and QMIX. However, these baselines often ignore the randomness in the situation. We propose MMD-MIX, a method that combines distributional reinforcement learning and value decomposition to alleviate the above weaknesses. Besides, to improve data sampling efficiency, we were inspired by REM (Random Ensemble Mixture) which is a robust RL algorithm to explicitly introduce randomness into the MMD-MIX. The experiments demonstrate that MMD-MIX outperforms prior baselines in the StarCraft Multi-Agent Challenge (SMAC) environment., 7 pages, 2 figures, 2 tables. Accepted by IJCNN 2021 more...

Human motion analysis is integral to many applications ranging from biomedicine to surveillance and 3D animation. The availability of consumer RGB-D sensors makes motion capture (Mocap) more prevalent and accessible in our daily lives. However, depth-based Mocap (D-Mocap) suffers significant errors and noise due to the limitation of depth sensing, selfocclusion, and many other problems. We present a novel filter-assisted deep learning approach to improve low-quality human motion data (e.g., D-Mocap) by taking advantage of the recent progress in both deep learning and nonlinear Kalman filtering. At the heart of this method is a learned motion manifold through the use of a convolutional autoencoder trained on high-quality, richvariety CMU Mocap data which is used to recover valid human motion from corrupted input. Furthermore, the Tobit Kalman filter (TKF), proposed to handle censored measurements, is used to assist the autoencoder with more kinematic and dynamic constraints. Two structural paradigms are investigated to handle different kinds of data errors by maximizing the synergy between the two integral parts in this work. The experimental results on both simulated and real-world human motion data demonstrate the effectiveness and robustness of the proposed methods to improve the quality of noisy and erroneous Mocap data. more...

Increased accuracy and affordability of depth sensors such as Kinect has created a great depth-data source for various 3D oriented applications. Specifically, 3D model retrieval is attracting attention in the field of computer vision and pattern recognition due to its numerous applications. A cross-domain retrieval approach such as depth image based 3D model retrieval has the challenges of occlusion, noise and view variability present in both query and training data. In this paper, we propose a new supervised deep autoencoder approach followed by semantic modeling to retrieve 3D shapes based on depth images. The key novelty is the two-fold feature abstraction to cope with the incompleteness and ambiguity present in the depth images. First, we develop a supervised autoencoder to extract robust features from both real depth images and synthetic ones rendered from 3D models, which are intended to balance reconstruction and classification capabilities of mix-domain data. Then semantic modeling of the supervised autoencoder features offers the next level of abstraction to cope with the incompleteness and ambiguity of the depth data. It is interesting that unlike any other pairwise model structures, we argue that cross-domain retrieval is still possible using only one single deep network trained on real and synthetic data. The experimental results on the NYUD2 and ModelNet10 datasets demonstrate that the proposed supervised method outperforms the recent approaches for cross-modal 3D model retrieval. more...

Objectives Our goal was to investigate risk factors for acute kidney injury (AKI) after coronary artery bypass grafting (CABG) and the impact of AKI on short-term outcomes. Methods Data on 1395 patients (1261 who had isolated CABG and 134 with other operations) who underwent non-emergent CABG from January 2013 to March 2016 were retrospectively collected from a single centre. Logistic regression was performed to analyse risk factors. Cox regression was used to analyse the impact of AKI on the postoperative 30-day death rate. A 1:1 propensity score matching was performed to balance the baseline characteristics. Results The incidence of AKI with on-pump and off-pump coronary artery bypass was 10.4% and 3.5%, respectively. With logistic regression, duration of surgery was a risk factor for AKI (stage ≥2); previous hypertension, preoperative renal function insufficiency and the presence of cardiopulmonary bypass (CPB) were risk factors for mild AKI (stage ≥1). CPB time >207.5 min could be used to predict AKI (sensitivity 79.2%, specificity 78.6%) in the combined group. After adjusting for the duration of the operation, postoperative AKI (stage ≥1) was a risk factor for 30-day death and there was no difference in the 30-day death rate between on-pump and off-pump CABG. Conclusions The use of CPB was a risk factor for mild AKI that did not affect the 30-day death rate of CABG whereas moderate to severe AKI caused by prolonged CPB time associated with surgical complexity affected the 30-day death rate. AKI may indicate surgical injury. The decision to use the on- or off-pump technique does not affect the 30-day death rate of CABG. more...

This paper addresses a terminal sliding mode control U+0028 T-SMC U+0029 method for load frequency control U+0028 LFC U+0029 in renewable power systems with generation rate constraints U+0028 GRC U+0029. A two-area interconnected power system with wind turbines is taken into account for simulation studies. The terminal sliding mode controllers are assigned in each area to achieve the LFC goal. The increasing complexity of the nonlinear power system aggravates the effects of system uncertainties. Radial basis function neural networks U+0028 RBF NNs U+0029 are designed to approximate the entire uncertainties. The terminal sliding mode controllers and the RBF NNs work in parallel to solve the LFC problem for the renewable power system. Some simulation results illustrate the feasibility and validity of the presented scheme. more...

Single image dehazing algorithms are recently attracting more and more attention from many researchers because of their flexibility and practicality. However, most existing algorithms have some challenges in dealing with images captured under complex weather conditions because the often used assumptions cannot always reflect true structural information of natural images in those situations. In this paper, we develop a deep transmission network to estimate the transmission map for efficient image dehazing, which automatically explores and exploits underlying haze-relevant features from RGB color channels and a local patch jointly for robust transmission estimation. Moreover, due to the fact that transmission values are affected by light wavelengths, a three-channel transmission map is considered in the proposed network so that this network can discover and utilize the chromatic characteristics for transmission estimation. We also investigate different network structures and parameter settings to achieve different trade-offs between performance and speed, and find that three color channels and local spatial information are the most informative haze-relevant features. This could explain why haze relevant priors or assumptions are often related to three color channels in most existing methods. Experiment results demonstrate that the proposed algorithm outperforms state-of-the-art methods on both synthetic and real-world datasets. more...

The aim of the present study was to investigate the association between social psychological factors and the occurrence of laryngeal cancer. A 1:1 matched case-control study was conducted. The participants completed questionnaires that included general information, such as the Life Event Scale, the Social Support Rating Scale, and the Eysenck Personality Questionnaire (EPQ). Scores were compared between the groups using paired t-tests and Wilcoxon's signed-rank tests. No significant difference in the psychoticism scale of the EPQ was observed between the two groups (P>0.05). However, significant differences were observed in scores on the life events and social support scales and in the remaining dimensions of the EPQ (all P-values more...

We study two fundamental issues about video-based human walking estimation, where the goal is to estimate 3D gait kinematics (i.e., joint positions) from 2D gait appearances (i.e., silhouettes). One is how to model the gait kinematics from different walking styles, and the other is how to represent the gait appearances captured under different views and from individuals of distinct walking styles and body shapes. Our research is conducted in three steps. First, we propose the idea of joint gait-pose manifold (JGPM), which represents gait kinematics by coupling two nonlinear variables, pose (a specific walking stage) and gait (a particular walking style) in a unified latent space. We extend the Gaussian process latent variable model (GPLVM) for JGPM learning, where two heuristic topological priors, a torus and a cylinder, are considered and several JGPMs of different degrees of freedom (DoFs) are introduced for comparative analysis. Second, we develop a validation technique and a series of benchmark tests to evaluate multiple JGPMs and recent GPLVMs in terms of their performance for gait motion modeling. It is shown that the toroidal prior is slightly better than the cylindrical one, and the JGPM of 4 DoFs that balances the toroidal prior with the intrinsic data structure achieves the best performance. Third, a JGPM-based visual gait generative model (JGPM-VGGM) is developed, where JGPM plays a central role to bridge the gap between the gait appearances and the gait kinematics. Our proposed JGPM-VGGM is learned from Carnegie Mellon University MoCap data and tested on the HumanEva-I and HumanEva-II data sets. Our experimental results demonstrate the effectiveness and competitiveness of our algorithms compared with existing algorithms. more...

In this paper, we propose a new asymmetric supervised deep autoencoder approach to retrieve 3D shapes based on depth images. The asymmetric supervised autoencoder is trained with real and synthetic depth images together. The novelty of this research lies in the asymmetric structure of a supervised deep autoencoder. The proposed asymmetric deep supervised autoencoder deals with the incompleteness and ambiguity present in the depth images by balancing reconstruction and classification capabilities in a unified way with mixed depth images. We investigate the relationship between the encoder layers and decoder layers, and claim that an asymmetric structure of a supervised deep autoencoder reduces the chance of overfitting by 8% and is capable of extracting more robust features with respect to the variance of input than that of a symmetric structure. The experimental results on the NYUD2 and ModelNet10 datasets demonstrate that the proposed supervised method outperforms the recent approaches for cross modal 3D model retrieval. more...

We present an online framework for dense 3D reconstruction of indoor scenes using sequential Manhattan keyframes. We take advantage of the global geometry of indoor scenes extracted by Manhattan frames and pose optimization to enhance the accuracy and robustness of the reconstructed models. During sequential reconstruction, a Manhattan frame is extracted for each keyframe after surface normal adjustment, and used for a Manhattan keyframe-based planar alignment to initialize the surface registration while a pose graph optimization is used to refine camera poses. The final model is created by integrating the Manhattan keyframes into the unified volumetric model using refined pose estimations. Experimental results demonstrate the advantage of our geometry-based approach to reduce the cumulative registration error and overall geometric drift. more...

There are two common paradigms for indoor scene understanding, pixel-level labeling and bounding box generation. The two tasks have a complementary nature but are normally achieved separately with different computational flows. We propose a novel method to bridge the two tasks by creating category-specific 3D bounding box hypotheses from score-maps of any deep networks trained on pixel-level semantic labels along with depth data. Those hypotheses can be further used to locate all objects as different non-overlapping bounding boxes by incorporating high-level knowledge, such as common room settings, co-existence or co-exclusiveness etc. We develop an objective function that involves confidence scores and the depth visibility to initialize and optimize multiple hypotheses for each category-specific score map. Experiment results show that our method significantly outperforms direct bounding box generation using pixel-level labeling. more...

An autoencoder is a class of neural network that is trained to output an accurate reproduction of the input while learning key lower dimensional features, otherwise known as a manifold. A lower dimensional representation of the original input, referred to as the latent space, encodes the intrinsic data structure over the manifold. This paper proposes filter-guided manifold optimization in the latent space of a convolutional autoencoder to recover noisy motion data collected by a depth sensor. Autoencoder output is smoothed using four traditional filters and employed as target motion data in an objective function. The difference between the actual output and target is minimized through stochastic gradient descent over the latent space, using manifold optimization to produce the expected smooth output. The advantage of this filter-guided approach over traditional filtering is that the resultant motion data still adheres to the manifold in the latent space learned by the autoencorder from training on motion data. more...

In this paper, we propose a three-stream convolutional neural network (3SCNN) for action recognition from skeleton sequences, which aims to thoroughly and fully exploit the skeleton data by extracting, learning, fusing and inferring multiple motion-related features, including 3D joint positions and joint displacements across adjacent frames as well as oriented bone segments. The proposed 3SCNN involves three sequential stages. The first stage enriches three independently extracted features by co-occurrence feature learning. The second stage involves multi-channel pairwise fusion to take advantage of the complementary and diverse nature among three features. The third stage is a multi-task and ensemble learning network to further improve the generalization ability of 3SCNN. Experimental results on the standard dataset show the effectiveness of our proposed multi-stream feature learning, fusion and inference method for skeleton-based 3D action recognition. more...

We propose an efficient approach for robust reconstruction of indoor scenes by taking advantage of the geometric relation between consecutive Manhattan keyframes and local pose refinement to improve the accuracy and fidelity of the reconstructed models. At the core of our framework, we have a Local Manhattan Frame Growing system, which finds the principal directions of the scene and aligns point clouds with the dominant plane, and a Local Pose Optimization, which refines the pose estimation for a specific range of frames. During the reconstruction process, we use Manhattan keyframes for a planar pre-alignment to provide a robust initialization for the final surface registration. All Manhattan keyframes are integrated using a frame-to-model scheme to create local models based on the refined camera poses. The final dense model is reconstructed by adopting a geometric registration between local segments and integrating them into a global frame. The experimental results show the effectiveness of our approach to reduce the cumulative registration error and overall geometric drift. more...

There are three levels for indoor scene understanding, pixel level labeling, object level recognition and scene level holistic understanding. The three levels provide complementary bottom-up scene representation. Traditional research often addresses these three tasks separately where the three levels of semantic data are seldom jointly considered. We propose a new method to bridge the three semantic levels by using dual graphical models for relational modeling of object categories in indoor scenes. The vertical placement model captures top-down object configuration by which the visible pixels of some accessory objects could be used to infer the presence of a supportive object underneath. The horizontal placement model reveals how multiple object categories are related to each other on the ground in different indoor scenes. The experimental results show improvements on the bounding box accuracy using both vertical and horizontal placement models from pixel level labeling. more...

A P-Band high gain radio frequency (RF) amplifier is presented in this paper. It is fabricated in a 0.35 μm silicon-germanium (SiGe) heterojunction bipolar transistor (HBT) technology. Theory, simulation and measurement are shown. It is designed as a two-stage cascade amplifier. The input and output matching networks are designed on the chip. It achieves high power gain of 29.5 dB at 500 MHz, wide bandwidth of 1 GHz, low noise figure of 1.1 dB at 500 MHz, and high output 1 dB compression point of 11 dBm at 500 MHz. It is designed for narrow or wide bandwidth industrial and military applications that require high gain and low noise IF or RF amplification. It can also be easily integrated into a RF system-on-a-chip (SOC). more...

A high input impedance operational amplifier (OPA) with bipolar junction field effect transistor (Bi-JFET) technology is presented. By using JFET differential pair, it achieves very low input bias offset current (pA magnitude) and low noise (3 nV/sqrt(Hz)). Moreover, bipolar gain stage and output stage provides high open-loop gain (120 dB), high output voltage slew rate (more than 10 V/μs) and broadband performance (more than 10 MHz). The proposed OPA is very suitable for detecting and amplifying all kinds of weak signals. more...

Contrast enhancement which aims to increase the contrast of an image with low dynamic range, has been widely studied and exploited. In spite of the great success of many contrast enhancement algorithms, they still have difficulty in achieving both global and local contrast enhancement so that some over-enhancement, under-enhancement or even halo artifacts are often produced in complex images. This paper proposes a simple but efficient contrast enhancement method which may achieve both global and local contrast enhancement and perceptually suppress the above-mentioned problems. A cost function integrating global enhancement with local contrast enhancement is firstly constructed to pose image enhancement as an optimization problem. Then, two key steps are involved in solving for an optimal solution, the just-noticeable difference (JND) model is introduced to perceptually determine maximum local gains and neighbouring gray-level difference for local and global contrast enhancement, respectively, and an adaptive parameter regularization method is invoked to further suppress over-enhancement and halo artifacts. The experimental results on many images both qualitatively and quantitatively demonstrate our algorithm can robustly provide better visual quality in global and local contrast compared to a selection of other well-known state-of-the-art algorithms. more...

Single image dehazing has captured much attention due to increasing applications. However, state-of-the-art image dehazing algorithms often suffer from undesirable quantization artifacts and noises in heavily hazy regions or sky patches of hazy image where dense scattering often occurs, so that dehazed results may have poor image quality or even lose the original spectral or structural information. To address this problem, we propose a perception oriented transmission estimation method for high quality image dehazing. As the key contribution, a novel transmission model is firstly developed by posing image dehazing as a local contrast optimization problem. This transmission model can flexibly adjust haze removal to accommodate the expected local contrast gain. Specially, this model can lead to a solution which is similar to the one using the dark channel prior, but it is not confined to the dark channel prior assumption. Then, in order to remove haze and simultaneously suppress quantization artifacts and noises, two specific steps are introduced. First, we develop a scattering-aware method via a Bayesian framework to estimate the scattering probability of each pixel in a hazy image. Second, a perceptually adaptive parameter selection scheme is proposed to determine the expected contrast gain for the transmission estimation by taking advantage of the just-noticeable-distortion (JND) model. Experimental results demonstrate that the proposed algorithm can effectively remove haze and suppress undesirable degradation on dehazed images, both quantitatively and qualitatively, when compared with the state-of-the-art algorithms under dense scattering conditions. more...

A new longitudinal attitude control system design for an unmanned seaplane in the severe sea states is presented in this paper. We develop a nonlinear passive observer, which is used to achieve wave filtering and state estimation. Moreover, the observer can be extended to achieve adaptive wave filtering in varying sea states. Using the estimated low-frequency states, a backstepping sliding mode controller is designed to keep the longitudinal attitude of the unmanned seaplane stable. The stability of the total closed loop system is analyzed by using Lyapunov theory. Simulations are performed in different wave conditions, including Seastate 3 and Seastate 5. The simulations results show that the proposed longitudinal attitude controller can improve the anti-waves capability effectively. Moreover, adaptive wave filter has a significant advantage over a filter with fixed model parameters in varying sea states. more...

Dividing wall column (DWC) is a promising process intensification technology of distillation. Researchers have been investigating different control structures for different DWC configurations. However, research on control of the four-product Kaibel column is relatively scarce. The focus of this paper is to investigate different control structures of the four-product Kaibel column. The separation of n-pentane, n-hexane, n-heptane and n-octane is used as a case study of control structure selection in this paper. Three control structures are studied: control structure with fixed split ratios (CS1), control structure with an active liquid split (CS2) and control structure with an active vapor split (CS3). CS2 and CS3 are able to stabilize the process and let the product compositions return near the set points. The dynamic performances of the last two proposed control structures show that they are able to handle feed disturbances. This work demonstrates that traditional single-loop PID control is able to handle disturbances inserted into the four-product Kaibel (dividing wall) column. more...

The key to feature fusion for classification is to take advantage of the discriminative and complementary information from different feature sets, which can be represented as internal (within each feature set) or external structures (across different feature sets). Traditional approaches tend to preserve either internal or external structures via certain feature projection. Some early attempts consider both structures implicitly or indirectly without revealing their relative importance and relevance in feature fusion. We propose a new unsupervised heterogeneous structure fusion (HSF) algorithm that is able to jointly optimize two kinds of structures explicitly and directly via unified feature projection. Specifically, the internal structure is represented based on Locality Preserving Projection (LPP), and the external structure is captured by Canonical Correlation Analysis (CCA). The objective function of HSF combines two feature structures in a closed form which can be optimized alternately via linear programming and eigenvector methods. The HSF solution provides not only the optimal feature projection but also the weights that encode the relative importance between two kinds of feature structures. The experimental results on image classification, face recognition, shape analysis and infrared target recognition demonstrate the effectiveness and efficiency of HSF compared with state-of-the-art methods. Graphical abstractDisplay Omitted HighlightsThe external and internal feature structures are jointly optimized for fusion.Two kinds of feature structures are preserved via a unified feature projection.The nonlinear relationship between two kinds of feature structures is revealed.The new feature fusion algorithm shows great promise in four classification problems. more...

This paper proposed a fast facedetection method based on the skin color feature and local binary gradient feature. First, according to the clustering of human skin color in the YCbCr color space, the skin color area is detected in the image. Then, it is coarsely fast judged whether if the face is in the skin color area. Finally, the local binary gradient feature is used to judge face accurately, and the weights of the local binary gradient feature is found by using AdaBoost train algorithm. For improved the efficiency of algorithm, the algorithm is accelerated by using the method of the integral image, the cascade classifier and the search order from big to small. The algorithm was tested by a set with 450 color images which the size is 896 ×592. It can find that the average detection time of new algorithm has reduced about 17.1% compare with the face detection algorithm of Paul Viola. The detection accuracy is similar with algorithm of Paul Viola. more...

A new output feedback control design for robust velocity and altitude tracking of an air-breathing hypersonic vehicle (AHSV) is presented in this paper. The control scheme is performed on the assumption that only partial states of AHSV are measurable. The key idea is to employ the immersion and invariance approach to design globally asymptotically stable observers for the unmeasurable states. For controller design, the whole control architecture is constructed using dynamic surface control, based on the decomposition of the longitudinal dynamics of AHSV into velocity and altitude subsystems. Stability analysis is presented using the Lyapunov theory. Representative simulations are carried out on the high-fidelity model, which illustrate the effectiveness and robustness of the proposed scheme. more...

We propose a generative framework for 3D human pose estimation that is able to operate on both individual point sets and sequential depth data. We formulate human pose estimation as a point set registration problem, where we propose three new approaches to address several major technical challenges in this research. First, we integrate two registration techniques that have a complementary nature to cope with non-rigid and articulated deformations of the human body under a variety of poses. This unique combination allows us to handle point sets of complex body motion and large pose variation without any initial conditions, as required by most existing approaches. Second, we introduce an efficient pose tracking strategy to deal with sequential depth data, where the major challenge is the incomplete data due to self-occlusions and view changes. We introduce a visible point extraction method to initialize a new template for the current frame from the previous frame, which effectively reduces the ambiguity and uncertainty during registration. Third, to support robust and stable pose tracking, we develop a segment volume validation technique to detect tracking failures and to re-initialize pose registration if needed. The experimental results on both benchmark 3D laser scan and depth datasets demonstrate the effectiveness of the proposed framework when compared with state-of-the-art algorithms. more...

Obstacle detection in urban traffic is a hot topic in intelligent visual surveillance systems. In this paper, a real-time automatic obstacle recognition method based on computer vision technology is presented. The proposed method aims at detecting and recognizing the road obstacles such as abandoned objects, accident vehicles and illegally parked vehicles, which can prevent the traffic accident effectively. In the method, the target images are captured by a visible image sensor firstly. In order to avoid the static objects disappearing from foreground in short time when using GMM (Gaussian Mixture Model), background is built and foreground objects are extracted by the proposed algorithm SUOG (Selective Updating of GMM). Relative object speed is used to detect the static obstacles, and FROI (Flushed Region of Interest) algorithm based on the concept of connected domain, is presented to eliminate noises outside road and improve real-time capability. At last, a classification method of adaptive interested region based on HOG and SVM, and a new recognition algorithm of accident vehicles based on multi-feature fusion are proposed to classify the road obstacles. Experiments indicate that the detection rate of the proposed obstacle detection method is up to 96 % in urban road traffic. Through experiment, it is shown that the developed obstacle detection method has low computational complexity, and can fulfill the requirement of real-time applications, and it is correct and effective. more...

Depth image based 3D model retrieval faces challenges of occlusion, noise, and view variability present in depth images. In this work, we study a new supervised deep autoencoder for depth image-based 3D model retrieval. We investigate both supervised and unsupervised approaches to bring synthetic depth images rendered from 3D models and real depth images in the same feature space. We show that providing appropriate supervision in back propagation of the autoencoder can help the retrieval performance. The key novelty is the new objective function where supervised classification information is combined with the reconstruction error for joint optimization. It is interesting to manifest that, unlike any other pairwise model structures, crossdomain retrieval is still possible using only one single deep network in our model. We have evaluated the effectiveness of our model on NYUD2 depth image dataset and Model-Net10 models for ten indoor object categories. We have rendered 95 different views for each 3D model and found that training rendered and real depth images together is an effective way to bridge the gap between 3D models and depth data. The proposed supervised method outperforms the recent pairwise approach. more...

A novel class AB two stage operational amplifier was presented to deal with the problem of the conventional class A two stage operational amplifier’s limited slew rate. The output stage quiescent current was controlled accurately and the slew rate and unity gain bandwidth were enhanced significantly, along with no additional static power dissipation. Simulation results show that the positive and negative slew rates were raised to 35 and 2.2 times and unity gain bandwidth was raised to 5 times when compared with conventional class A operational amplifier. more...

We propose an efficient 3D modeling method to support real-time volumetric reconstruction of indoor scenes based on sequential depth sequences captured from a RGB-D camera. Specifically, we want to reduce the cumulative error from sequential ICP registration due to noise and outliers in the depth data. We take advantage of the Manhattan frame assumption valid in most indoor scenes that can be used to facilitate large scale 3D surface registration. In our approach, the Manhattan frame is extracted from each depth frame and used for plane-to-plane frame alignment to initialize point-to-plane ICP surface registration. Experimental results on three different indoor datasets including LIDAR ground-truth data demonstrate the advantages of the proposed algorithm over the original ICP-based approaches to volumetric reconstruction. more...

We propose a new cuboid matching algorithm for robust object detection in RGB-D images from an indoor scene. Unlike traditional bounding boxes, a cuboid is more flexible and accurate to represent the object's orientation and basic geometry that can serve as an informative mid-level representation for scene understanding. However, over-detection and miss-detection are common problems when the scene is too cluttered and has many irrelevant planar surfaces. We approach these problems from two perspectives. First, we apply a few planar features to improve initial plane generation and to select dominant plane candidates for cuboid initialization. Second, cuboid candidates are optimized to their local fitness involving both color and depth features. The experimental results show significant improvements over the state-of-art method both quantitatively and qualitatively. more...

Rare class objects in natural images often convey semanti-cally important information than background for scene understanding, but they are often overlooked during image parsing due to their low occurrence frequency and limited spatial coverage. In this work, we present a superpixel-based and rare class-oriented scene labeling framework (sRCSL), which seamlessly incorporates edge features into an integration of global and local CNN models. A dual-mode coarse-to-fine superpixel representation is developed for accurate yet efficient labeling, where coarse and fine superpixels are applied to background and rare classes respectively. Furthermore, saliency detection is incorporated by the combination of probabilistic belief maps from local and global inference. Experimental results demonstrate promising performance of the proposed framework on the SIFTflow dataset both qualitatively and quantitatively for semantic labeling, especially for rare classes. more...

Flexible air-breathing hypersonic vehicles feature significant uncertainties which pose huge challenges to robust controller designs. In this paper, four major categories of uncertainties are analyzed, that is, uncertainties associated with flexible effects, aerodynamic parameter variations, external environmental disturbances, and control-oriented modeling errors. A uniform nonlinear uncertainty model is explored for the first three uncertainties which lumps all uncertainties together and consequently is beneficial for controller synthesis. The fourth uncertainty is additionally considered in stability analysis. Based on these analyses, the starting point of the control design is to decompose the vehicle dynamics into five functional subsystems. Then a robust trajectory linearization control (TLC) scheme consisting of five robust subsystem controllers is proposed. In each subsystem controller, TLC is combined with the extended state observer (ESO) technique for uncertainty compensation. The stability of the overall closed-loop system with the four aforementioned uncertainties and additional singular perturbations is analyzed. Particularly, the stability of nonlinear ESO is also discussed from a Lienard system perspective. At last, simulations demonstrate the great control performance and the uncertainty rejection ability of the robust scheme. more...

Unmanned seaplanes are special kinds of fixed-wing airplanes which can achieve autonomous takeoff and landing on water. Due to complicated hydrodynamic forces and unpredictable sea states, researches on modeling, dynamic analysis and controller design are still facing great challenges. In this paper, the dynamic characteristics and motion stability of the unmanned seaplane are firstly analyzed based on a nonlinear mathematical model. The proposed autonomous takeoff control system consists of two main parts: T-S fuzzy identification and generalized predictive control (GPC). A linear CARIMA model obtained through T-S fuzzy identification is used to represent the dynamic characteristics in different motion stages. Wave forecasting is considered in the GPC algorithm to improve the anti-waves capability and avoid unstable phenomena in high sea states. Simulations are performed in three different wave conditions, including calm water, regular wave and irregular wave. Moreover, GPC controller is compared with other disturbance rejection control solutions. The simulation results show that the proposed controller has good performances for autonomous takeoff of the unmanned seaplane. more...

A new nonlinear adaptive control scheme based on the immersion and invariance theory is presented to achieve robust velocity and altitude tracking for hypersonic vehicles with parametric uncertainty. The longitudinal dynamics of the hypersonic vehicle are first decomposed into velocity, altitude/flight-path angle, and angle of attack/pitch rate subsystems. Then a non-certainty-equivalent controller based on immersion and invariance, consisting of a control module and a parameter estimator, is designed for each subsystem with all the aerodynamic parameters unknown. The main feature of this method lies in the construction of the estimator, which is a sum of a partial estimate generated from the update law and an additional nonlinear term. The new term is capable of assigning appointed stable dynamics to the parameter estimate error. Stability analysis is presented using Lyapunov theory and shows asymptotical convergence of the tracking error to zero. Representative simulations are performed. Rapid and accurate command tracking is demonstrated in these numerical simulations, which illustrate the effectiveness and robustness of the proposed approach. more...

We propose a new joint view-identity manifold (JVIM) for multi-view and multi-target shape modeling that is well-suited for automated target tracking and recognition (ATR) in infrared imagery. As a shape generative model, JVIM features a novel manifold structure that imposes a conditional dependency between the two shape-related factors, view and identity, in a unified latent space, which is embedded with one view-independent identity manifold and infinite identity-dependent view manifolds. A modified local linear Gaussian process latent variable model (LL-GPLVM) is proposed for JVIM learning where a stochastic gradient descent method is used to improve the learning efficiency. We also develop a local inference technique to speed up JVIM-based shape interpolation. Due to its probabilistic and continuous nature, JVIM provides effective shape synthesis and supports robust ATR inference for both known and unknown target types under arbitrary views. Experiments on both synthetic data and the SENSIAC infrared ATR database demonstrate the advantages of the proposed method over several existing techniques both qualitatively and quantitatively. more...

This paper presents a two-layer gait representation framework for video-based human motion estimation that extends our recent dual gait generative models, visual gait generative model (VGGM) and kinematic gait generative model (KGGM), with a new capability of part-whole gait modeling. Specifically, the idea of gait manifold learning is revisited to capture the gait variability among different individuals at both whole and part levels. A key issue is the selection of an appropriate distance metric to evaluate the dissimilarity between two gaits (either at whole or part levels) that determines an optimal manifold topology. Several metrics are studied and compared in terms of their effectiveness for gait manifold learning at both whole and part levels. This work involves one whole-based and two part-level gait manifolds by which three pairs of KGGM and VGGM can be learned and integrated for part-whole gait modeling. Moreover, a two-stage Monte Carlo Markov Chain (MCMC) inference algorithm is developed for video-based part-whole motion estimation. The proposed algorithm is tested on the HumanEva data and reaches state-of-art results. more...

In this study, a robust control scheme based on dynamic inversion using fuzzy neural networks (FNN) is designed to control a nonlinear aircraft driven by two different actuators (fins and reaction jets). First, the dynamic model of the aircraft with reaction jets is briefly introduced. Then, the robust control scheme is explained in detail, in which the dynamic inversion is adopted to solve the system nonlinearity and the FNN cancels model uncertainties by on-line learning. Learning is accomplished by a simple weight update rule derived from Lyapunov theory, thus assuring the stability of the closed loop system. In addition, a control moment allocator is designed to coordinate the work of fins and reaction jets. Finally, the control scheme is demonstrated in simulation. Its robustness is evaluated relative to a PID control based on dynamic inversion; its rapid time response performance is compared with a conventional aircraft with fins only, to certify the effectiveness of the control moment allocator. more...

A robust trajectory linearization control (TLC) scheme is presented for a generic hypersonic vehicle (GHV) entry flight. The basic TLC frame constructs a baseline controller for the GHV attitude system, providing local closed-loop exponential stability along nominal trajectories. Then two strategies are integrated with the basic TLC for robustness enhancement. For one thing, to cope with diverse perturbations, extended state observer is designed to yield a compensation control law. For the other, an adaptive time-varying bandwidth algorithm is developed, which can not only avoid actuator saturation and integrator windup, but most importantly, also improve system robustness when huge dynamic pressure variation occurs during entry flight. This integrated robust TLC scheme can guarantee a better control performance and a larger stability domain. At last, the great advantages of the proposed robust TLC scheme is demonstrated via three groups of simulation, including a three-DOF attitude tracking, a BTT-180 maneuver simulation, and a six-DOF integrated guidance and control test. more...

State-of-the-art single image dehazing algorithms have some challenges to deal with images captured under complex weather conditions because their assumptions usually do not hold in those situations. In this paper, we develop a deep transmission network for robust single image dehazing. This deep transmission network simultaneously copes with three color channels and local patch information to automatically explore and exploit haze-relevant features in a learning framework. We further explore different network structures and parameter settings to achieve tradeoffs between performance and speed, which shows that color channels information is the most useful haze-relevant feature rather than local information. Experiment results demonstrate that the proposed algorithm outperforms state-of-the-art methods on both synthetic and real-world datasets. more...

A novel adaptive control design based on the immersion and invariance (I&I) theory is presented for the non-minimum phase hypersonic vehicles with parameter uncertainties. The vehicle dynamics are first divided into the velocity subsystem, the altitude/flight-path angle subsystem and the angle of attack/pitch rate subsystem. Then three I&I based adaptive controllers are designed for each subsystem respectively. The key feature of the proposed control scheme lies in the construction of the estimator, which can be used to assign prescribed dynamics to the estimate errors. The interaction of the flexible modes with the rigid dynamics is considered in both the synthesis of the controller and the analysis of the closed-loop system. A complete stability analysis of the closed-loop system is presented using Lyapunov theory. Simulation results show that the proposed control system provides precise and robust tracking of the velocity and altitude commands, as well as accurate regulation of angle of attack. more...

Face images under uncontrolled environments suffer from the changes of multiple factors such as camera view, illumination, expression, etc. Tensor analysis provides a way of analyzing the influence of different factors on facial variation. However, the TensorFace model creates a difficulty in representing the nonlinearity of view subspace. In this paper, to break this limitation, we present a view-manifold-based TensorFace (V-TensorFace), in which the latent view manifold preserves the local distances in the multiview face space. Moreover, a kernelized TensorFace (K-TensorFace) for multiview face recognition is proposed to preserve the structure of the latent manifold in the image space. Both methods provide a generative model that involves a continuous view manifold for unseen view representation. Most importantly, we propose a unified framework to generalize TensorFace, V-TensorFace, and K-TensorFace. Finally, an expectation-maximization like algorithm is developed to estimate the identity and view parameters iteratively for a face image of an unknown/unseen view. The experiment on the PIE database shows the effectiveness of the manifold construction method. Extensive comparison experiments on Weizmann and Oriental Face databases for multiview face recognition demonstrate the superiority of the proposed V- and K-TensorFace methods over the view-based principal component analysis and other state-of-the-art approaches for such purpose. more...

Underground mine surveillance can record the situation of coal production site timely and accurately. Because of the poor quality of air, inadequate lighting, lower monitor equipment resolution, surveillance images are blurred. This paper makes full use of the complementary information between surveillance images sequence. And we reconstruct high resolution images by using of maximum a posteriori (MAP) algorithm. Experiment results show that the reconstructed images can reflect the detail of the underground mine surveillance images. Compared with the traditional interpolation results, the reconstructed images avoid jagged effect and overall fuzzy. The PSNR of images are also improved. more...