Chung-Lin Huang

Gait analysis for human identification through manifold learning and HMM

Gait recognition is a process of identifying individuals by the way they walk. Gait is often used as a unobstrusive biometric offering the possibility to identify people at a distance without any interaction or co-operation with the subject. This paper presents a novel method for both automatic viewpoint and person identification using only the silhouette sequence of gait. The gait silhouettes are nonlinearly transformed into low dimensional embedding and the dynamics in time-series images are modeled by HMM in the corresponding embedding space. The experimental results demonstrate that the proposed algorithm is an encouraging progress for the gait analysis research.

Gait Analysis For Human Identification Through Manifold Learning and HMM

With the increasing demands of visual surveillance systems, human identification at a distance has gained more interest. Gait is often used as an unobtrusive biometric offering the possibility to identify individuals at a distance without any interaction or co-operation with the subject. This paper presents a novel effectively method for automatic viewpoint and person identification by using only the sequence of gait silhouette. The gait silhouettes are nonlinearly transformed into low dimensional embedding and the dynamics in time-series images are modeled by HMM in the corresponding embedding space. The experimental results demonstrate that the proposed algorithm is an encouraging progress for automatic human identification.

A multi-view vision-based hand motion capturing system

Vision-based hand motion capturing approaches play a critical role in human computer interface owing to its non-invasiveness, cost effectiveness, and user friendliness. This work presents a multi-view vision-based method to capture hand motion. A 3-D hand model with structural and kinematical constraints is developed to ensure that the proposed hand model behaves similar to an ordinary human hand. Human hand motion in a high degree of freedom space is estimated by developing a separable state based particle filtering (SSBPF) method to track the finger motion. By integrating different features, including silhouette, Chamfer distance, and depth map in different view angles, the proposed motion tracking system can capture the hand motion parameter effectively and solve the self-occlusion problem of the finger motion. Experimental results indicate that the hand joint angle estimation generates an average error of 11^o.

Abnormal Event Detection in Video Using N-cut Clustering

This paper introduces an unusual event detection scheme in various video scenes. The proposed method finds out the video clips that are most different from the others based on the similarity measure. Each video clip is represented by the motion magnitude and direction histograms and color histogram. Without searching key-frames, we calculate the similarity matrix by using \chi^2 difference or chamfer difference as the similarity measure of features in different clips. Finally, we apply n-cut clustering. Clusters with low self-similarity value are reported as unusual events.

People tracking and counting for applications in video surveillance system

This paper proposes a vision-based people counting system to count the number of persons entering or leaving the entrance of building. First, we develop a so-called adaptive AV codebook background model to segment foreground objects. In ROI (range of interest), we use the template matching method to find the objects, and then apply Hough transform to detect head contours to verify whether the object is a person. Second, we locate personal location and record the bottom center point as the trajectory point. We compute the color distance between the previous and current tracked object, and link the center points which belong the same person as the trajectory. Finally we analyze the trajectory to determine whether the person enters or exits or just passes by the entrance. The experimental results are illustrated to verify the system performance.

Gait analysis for human walking paths and identities recognition

In this paper, we propose a gait analysis method which extracts the dynamic and static information from human walking for walking path and identity recognition. First, we utilize the periodicity of swing distances to estimate the gait period for each gait sequence. For each gait cycle, we extract the dynamic information by analyzing the statistic histogram of motion vectors and static information using Fourier descriptors. The extracted information is transformed into lower dimensional embedding space to represent the subject. Given a test feature vector, the nearest neighbor classifier is applied to compare with the feature vectors in the gait database for human object identification. The proposed algorithm is evaluated on the CASIA gait database, and the experimental results demonstrate this new system achieves a high recognition rate.

Content-based adaptive media player for networked video

This paper proposes an adaptive media play-out (AMP) method based on perceived motion energy (PME). Most of current researches of AMP are focusing only on buffer control and packet scheduling. Here, we develop a PME-based threshold adjustment scheme for determining the play-out rate. Because human eyes are more sensitive to high motion videos, we avoid over-slowing the play-out rate in high motion video clips. We adjust the play-out based on the PME variation to improve the visual perceptibility.

Human Motion Parameter Capturing Using Particle Filter and Nonparametric Belief Propagation

We propose a motion capturing system for human walking in the side view. We build a 3D human model with structural and kinematical constraints and then use the particle filter (PF) and nonparametric belief propagation (NBP) for human tracking. To reduce the high-dimensional parameters, the separated particle filter for tracking six parts of human body is used. PF will estimate some initial pose, and then NBP will compute the results after several iterations. In the experiments, we show the estimated motion parameter of each frame. The error angle of our system is less than 11 degrees.

A multi-resolution image registration method for multimedia application

This paper proposes a multiresolution coarse-to-fine algorithm to align images. By using a hierarchical strategy, we can improve the execution speed and iterative convergence properties. Our optimization procedure uses the standard Levenberg-Marquardt method and the gradient of the image with respect to the alignment parameters to solve the general registration problem. Our algorithm is applied to align images undergoing affine, perspective and bilinear transformations. In the experiments, we show that our registration algorithm generates precise registration parameters.

Iris recognition using Dynamic Programming Matching Pursuit

In this paper, we propose a new method named dynamic programming based matching pursuit algorithm for iris-based personal identification. Based on the matching pursuit algorithm, it selects the most representative path to do iris recognition. Our system consists of identification and verification. Finally, we use the experimental results to demonstrate the efficacy of the proposed method and show that it attains a better ROC curve and faster speed than the conventional matching pursuit based iris recognition system.