Weiwei Xing

Crowd behaviors analysis and abnormal detection based on surveillance data

Crowd analysis and abnormal trajectories detection are hot topics in computer vision and pattern recognition. As more and more video monitoring equipments are installed in public places for public security and management, researches become urgent to learn the crowd behavior patterns through the trajectories obtained by the intelligent video surveillance technology. In this paper, the FCM (Fuzzy c-means) algorithm is adopted to cluster the source points and sink points of trajectories that are deemed as critical points into several groups, and then the trajectory clusters can be acquired. The feature information statistical histogram for each trajectory cluster which contains the motion information will be built after refining them with Hausdorff distances. Eventually, the local motion coherence between test trajectories and refined trajectory clusters will be used to judge whether they are abnormal. HighlightsA novel approach to crowd analysis and abnormal detection on trajectory data.Learn hidden structure information of an unstructured scene by FCM clustering.Get rough training trajectory clusters and refine with Hausdorff distances.Built feature information statistical histogram of each cluster as motion patterns.Compare test trajectory with its most possible belonging pattern for detecting anomaly.

Gait recognition based on joint distribution of motion angles

Gait as a biometric trait has the ability to be recognized in remote monitoring. In this article, a method based on joint distribution of motion angles is proposed for gait recognition. The new feature of the motion angles of lower limbs is defined and extracted from either 2D video database or 3D motion capture database, and the corresponding angles of right leg and left leg are joined together to work out the joint distribution spectrums. Based on the joint distribution of these angles, we build a feature histogram individually. In the stage of distance measurement, three types of distance vector are defined and utilized to measure the similarity between the histograms, and then a classifier is built to implement the classification. Experiments has been carried out both on CASIA Gait Database and CMU motion capture database, which show that our method can achieve a good recognition performance. We use a new feature extracted from lower limbs of a human for gait recognition.The feature is represented as joint distribution of motion angles for classification.Our method can work on both 2D and 3D data.Experimental results show our method can achieve higher accuracy compared with related work.Our method is also less affected by clothes of the moving person.

Hybrid motion graph for character motion synthesis

Objective: This paper proposes a novel framework of Hybrid Motion Graph (HMG) for creating character animations, which enhances the graph-based structural control by motion field representations for efficient motion synthesis of diverse and interactive character animations. Methods: In HMG framework, the motion template of each class is automatically derived from the training motions for capturing the general spatio-temporal characteristics of an entire motion class. Typical motion field for each class is then constructed. The smooth transitions among motion classes are then generated by interpolating the related motion templates with spacetime constraints. Finally, a hybrid motion graph is built by integrating the separate motion fields for each motion class into the global structural control of motion graph through smooth transition. Results: In motion synthesis stage, a character may freely switch among different motion classes in the hybrid motion graph via smooth transitions between motion templates and flow within each class through the continuous space of motion field with agile and the continuous control process. Conclusion: Experimental results show that our framework realizes the fast connectivity among different motion classes and high responsiveness and interactivity for creating realistic character animation of rich behaviors with limited motion data and computational resources.

Exposing image forgery by detecting traces of feather operation

Powerful digital image editing tools make it very easy to produce a perfect image forgery. The feather operation is necessary when tampering an image by copy-paste operation because it can help the boundary of pasted object to blend smoothly and unobtrusively with its surroundings. We propose a blind technique capable of detecting traces of feather operation to expose image forgeries. We model the feather operation, and the pixels of feather region will present similarity in their gradient phase angle and feather radius. An effectual scheme is designed to estimate each feather region pixels gradient phase angle and feather radius, and the pixels similarity to its neighbor pixels is defined and used to distinguish the feathered pixels from un-feathered pixels. The degree of image credibility is defined, and it is more acceptable to evaluate the reality of one image than just using a decision of YES or NO. Results of experiments on several forgeries demonstrate the effectiveness of the technique. A blind technique of detecting traces of feather operation image forgeries.Modeling influence of feather operation by gradient phase angle and radius.An approach to expose image forgeries region by traces of feather operation.The degree of credibility is defined to evaluate the reality of one image.

MGA for feature weight learning in SVM —a novel optimization method in pedestrian detection

Pedestrian detection is a challenging task in computer vision, which is often treated as classification problem of pattern recognition. However, dealing with the high dimensional features extracted from images, it turns out to be difficult to choose and combine the informative features for classification. In this paper, a novel optimization method—Metropolis based Genetic Algorithm (MGA) is proposed to solve this problem, and a novel pedestrian detector MGA-SVM is presented and implemented. In MGA, the metropolis criterion is adopted into GA for dynamical parents’ selection, which makes the algorithm get a stronger ability to jump out of local minimum as well as achieve convergence. To test the effectiveness of the proposed MGA, we implement it for feature weight learning in SVM pedestrian detector, which is named as MGA-SVM. The experimental results demonstrate the MGA has a better optimization capacity than original GA, which leads to a more accurate pedestrian detection result by using MGA-SVM.

Two-level superpixel and feedback based visual object tracking

Abstract While numerous superpixel-based tracking algorithms have been proposed and demonstrated successfully, there still remain some challenges, such as determining the number of superpixels, mining and exploiting the structural information of superpixels and handling the drifts. In this paper, we propose a tracking method with two-level superpixels and a novel update strategy based on feedback to deal with the challenges mentioned above. Firstly, Bilateral filter is introduced to filter out outliers and improve the boundary capability of object as well as segmentation of superpixels. Then two-level superpixel is proposed to determine superpixel number automatically through iterating instead of setting superpixel number empirically which affects the robustness of tracking algorithm. Moreover, a novel measuring method which considers color similarity and relative positions of superpixels is proposed to make a better use of structural information of superpixels and improve tracking performance by adding relative position of superpixels into the appearance model. Finally, a feedback based update strategy is presented to handle drifts existing in tracking by calculating the adaptation of appearance model and updating the parameters like superpixel number and relative position of superpixels. Experiments on challenging sequences and comparisons to state-of-the-art methods demonstrate the feasibility and effectiveness of the proposed tracking algorithm.

A rapid multi-source shortest path algorithm for interactive image segmentation

This paper addresses the problem of performing multiple objects, interactive image segmentation. Given a small number of pixels (seeds) with predefined labels, we can quickly and accurately determine the closest seed from each unlabeled pixel. By assigning each pixel to the label same as its closest seed, a rapid image segmentation result can be obtained. Since the shortest distance is considered in this paper which directs our attention of segmentation into path planning problem, Dijkstra occurs to mind. Modifying the classical single-source algorithm, a simple yet rapid multi-source Dijkstra (MSD) algorithm is put forward. From both theoretical and experimental aspects, the proposed algorithm performs quite well in resisting noise, and preserving the objects details. Moreover, under the situation of multiple sources, instead of performing Dijkstra several times to obtain the distance from each pixel to each seed and choose the closest seed, the proposed multi-source image segmentation algorithm could determine the closest seed by running Dijkstra only once. Its efficiency, which will not be affected by the number of initial seed settings, maintains the same as the Dijkstra.

A Novel Concurrent Generalized Deadlock Detection Algorithm in Distributed Systems

Detecting deadlocks has been considered an important problem in distributed systems. Many approaches are proposed to handle this issue; however, little attention has been paid on coordinating concurrent execution of distributed deadlock detection algorithms. Previous approaches may report incorrect results false negatives, and they are inefficient due to lack of proper coordination of concurrent execution. In this paper, we present a novel concurrent coordination algorithm for distributed generalized deadlock detection. The proposed algorithm aims to avoid false negatives and improve the performance when concurrently executing deadlock detection in a distributed system. Our algorithm adopts diffusion computation to distribute probe messages and employs priority-based method to coordinate concurrent algorithm instances. Priority carried in the received probe messages will be locally recorded by each initiator. Instead of being suspended by higher priority algorithm instances, lower priority algorithm instances can accomplish deadlock detection locally. The initiator with the highest priority will receive and collect all related resource requests information from lower priority instances in a hierarchical manner and perform global deadlock detection at last. We evaluate our algorithm on a bunch of event-driven simulations. The experimental results show that our approach can achieve better accuracy and efficiency compared to previous approaches.

Learning a Semantic Space by Deep Network for Cross-media Retrieval

With the growth of multimedia data, the prob- lem of cross-media (or cross-modal) retrieval has attracted considerable interest in the cross-media retrieval community. One of the solutions is to learn a common representation for multimedia data. In this paper, we propose a simple but effective deep learning method to address the cross-media retrieval problem between images and text documents for samples either with single or multiple labels. Specifically, two independent deep networks are learned to project the input feature vectors of images and text into an common (isomorphic) semantic space with high level abstraction (semantics). With the same dimensional feature representation in the learned common semantic space, the similarity between images and text documents can be directly measured. The correlation between two modalities is built according to their shared ground truth probability vector. To better bridge the gap between the images and the corresponding semantic concepts, an open-source CNN implementation called Deep Convolutional Activation Feature (DeCAF) is employed to extract input visual features for the proposed deep network. Extensive experiments on two publicly available multi-label datasets, NUS-WIDE and PASCAL VOC 2007, show that the proposed method achieves better results in cross-media retrieval compared with other state of the art methods.

A Symbolic Representation of Motion Capture Data for Behavioral Segmentation.

For building and understanding computational models of human motion, behavioral segmentation of human motion into actions is a crucial step, which plays an important part in many domains such as motion compression, motion classification and motion analysis. In this paper, we present a novel symbolic representation of human motion capture data, called the Behavior String (BS). Based on the BS, a further motion segmentation algorithm for human motion capture data is proposed. The human motion capture data is treated as a high-dimensional discrete data points, which are clustered by an alternative algorithm based on density, and each cluster is divided into a character. Then, the BS is produced for the motion data by temporal reverting. By analyzing the BS, the human motion capture data is segmented into distinct behavior segments and the cycles of motion are found. Experiments show that our method not only has a good performance in behavioral segmentation for motion capture data, but also finds cycles of motion and the motion clips of the same behaviors from long original motion sequence. Keywords-motion analysis; behavioral segmentation; clustering; motion capture data; cycle