A novel method for crowd flow modeling and anomaly detection is proposed for both coherent and incoherent scenes. The novelty is revealed in three aspects. First, it is a unique utilization of particle trajectories for modeling crowded scenes, in which we propose new and efficient representative trajectories for modeling arbitrarily complicated crowd flows. Second, chaotic dynamics are introduced into the crowd context to characterize complicated crowd motions by regulating a set of chaotic invariant features, which are reliably computed and used for detecting anomalies. Third, a probabilistic framework for anomaly detection and localization is formulated. The overall work-flow begins with particle advection based on optical flow. Then particle trajectories are clustered to obtain representative trajectories for a crowd flow. Next, the chaotic dynamics of all representative trajectories are extracted and quantified using chaotic invariants known as maximal Lyapunov exponent and correlation dimension. Probabilistic model is learned from these chaotic feature set, and finally, a maximum likelihood estimation criterion is adopted to identify a query video of a scene as normal or abnormal. Furthermore, an effective anomaly localization algorithm is designed to locate the position and size of an anomaly. Experiments are conducted on known crowd data set, and results show that our method achieves higher accuracy in anomaly detection and can effectively localize anomalies. 1. Introduction and Related

This work addresses the problem of human action recognition by introducing a representation of a human action as a collection of short trajectories that are extracted in areas of the scene with significant amount of visual activity. The trajectories are extracted by an auxiliary particle filtering tracking scheme that is initialized at points that are considered salient both in space and time. The spatiotemporal salient points are detected by measuring the variations in the information content of pixel neighborhoods in space and time. We implement an online background estimation algorithm in order to deal with inadequate localization of the salient points on the moving parts in the scene, and to improve the overall performance of the particle filter tracking scheme. We use a variant of the Longest Common Subsequence algorithm (LCSS) in order to compare different sets of trajectories corresponding to different actions. We use Relevance Vector Machines (RVM) in order to address the classification problem. We propose new kernels for use by the RVM, which are specifically tailored to the proposed representation of short trajectories. The basis of these kernels is the modified LCSS distance of the previous step. We present results on real image sequences from a small database depicting people performing 12 aerobic exercises.

Periodicity is at the core of the recognition of many actions. This paper takes the following steps to detect and measure periodicity. 1) We establish a conceptual framework of classifying periodicity in 10 essential cases, the most important of which are flashing (of a traffic light), pulsing (of an anemone), swinging (of wings), spinning (of a swimmer), turning (of a conductor), shuttling (of a brush), drifting (of an escalator) and thrusting (of a kangaroo). 2) We present an algorithm to detect all cases by the one and the same algorithm. It tracks the object independent of the object’s appearance, then performs probabilistic PCA and spectral analysis followed by detection and frequency measurement. The method shows good performance with fixed parameters for examples of all above cases assembled from the Internet. 3) Application of the method, completely unaltered, to a random half hour of CNN news has led to an 80 % score. 1.

Most tracking algorithms implicitly apply a coarse segmentation of each target object using a simple mask such as a rectangle or an ellipse. Although convenient, such coarse segmentation results in several problems in tracking—drift, switching of targets, poor target localization, to name a few—since it inherently includes extra non-target pixels if the mask is larger than the target or excludes some portion of target pixels if the mask is smaller than the target. In this paper, we propose a novel probabilistic framework for jointly solving segmentation and tracking. Starting from a joint Gaussian distribution over all the pixels, candidate target locations are evaluated by first computing a pixel-level segmentation and then explicitly including this segmentation in the probability model. The segmentation is also used to incrementally update the probability model based on a modified probabilistic principal component analysis (PPCA). Our experimental results show that the proposed method of explicitly considering pixellevel segmentation as a part of solving the tracking problem significantly improves the robustness and performance of tracking compared to other state-of-the-art trackers, particularly for tracking multiple overlapping targets. 1.

Abstract. We present a novel algorithm for approximating the parameters of a multivariate t-distribution. At the expense of a slightly decreased accuracy in the estimates, the proposed algorithm is significantly faster and easier to implement compared to the maximum likelihood estimates computed using the expectation-maximization algorithm. The formulation of the proposed algorithm also provides theoretical guidance for solving problems that are intractable with the maximum likelihood equations. In particular, we show how the proposed algorithm can be modified to give an incremental solution for fast online parameter estimation. Finally, we validate the effectiveness of the proposed algorithm by using the approximated t-distribution as a drop in replacement for the conventional Gaussian distribution in two computer vision applications: object recognition and tracking. In both cases the t-distribution gives better performance with no increase in computation. 1

Abstract. Tracking multi-object under occlusion is a challenging task. When occlusion happens, only the visible part of occluded object can provide reliable information for the matching. In conventional algorithms, the deducing of the occlusion relationship is needed to derive the visible part. However deducing the occlusion relationship is difficult. The interdetermined effect between the occlusion relationship and the tracking results will degenerate the tracking performance, and even lead to the tracking failure. In this paper, we propose a novel framework to track multi-object with occlusion handling according to sparse reconstruction. The matching with ℓ1-regularized sparse reconstruction can automatically focus on the visible part of the occluded object, and thus exclude the need of deducing the occlusion relationship. The tracking is simplified into a joint Bayesian inference problem. We compare our algorithm with the state-of-the-art algorithms. The experimental results show the superiority of our algorithm over other competing algorithms. 1

Abstract — In this paper, we propose online-learned classifiers for data association in multitarget tracking. The classifiers are dynamically constructed and incrementally online learned using image patches, which are associated based on location proimity. A biological inspired architecture is used to compute the classification label of image patch. The extracted image patches are coded and learned by a 3-layer neural network that implements in-place learning. We employ minimum-cost network flow optimization to associate tracks with the image patches based on their appearance and location proximities. The presented framework is applied to learn 11 targets encountered in a PETS2009 data set. Cross validation results show that the overall recognition accuracy is above 93%. The comparison with other learning algorithms is promising. The results of the implemented multitarget tracker demonstrate the effectiveness of the approach. Key Workds: Intelligent video surveillance system, object learning, and biologically inspired neural network.

To address the challenges on non-cooperative longdistance human identification and verification, we propose an innovative cost-efficient system for automatic long-range biometric recognition of noncooperative individuals in 24/7 operations. The system has three cameras. One is a wide field of view (WFOV) CCD video camera with an Infrared (IR) filter and powerful IR illuminators for human scan in a wide area at a long distance. The other two cameras are high resolution video cameras with narrow field of view (NFOV) and an IR filter & illuminators, mounted on a pan-tilt-unit (PTU) to capture the frontal view of human face and iris respectively. Once the frontal views of moving individuals are captured by the NFOV cameras, the face/iris models will be extracted and classified by the state-of-the-art face/iris recognizers. The hardware of the biometric system also includes one FPGA, three DSP processors, and one Zigbee module for fast bio-data analysis and wireless data transmission. 1.

Abstract Human eyes are highly efficient devices for scanning through a large quantity of low-level visual sensory data and delivering selective information to one’s brain for high-level semantic interpretation and gaining situational awareness. Over the last few decades, the computer vision community has endeavoured to bring about similar perceptual capabilities to artificial visual sensors. Substantial efforts have been made towards understanding static images of individual objects and the corresponding processes in the human visual system. This endeavour is intensified further by the need for understanding a massive quantity of video data, with the aim to comprehend multiple entities not only within a single image but also over time across multiple video frames for understanding their spatio-temporal relations. A significant application of video analysis and understanding is intelligent surveillance, which aims to interpret automatically human activity and detect unusual events that could pose a threat to public security and safety. 1