Yang Tao

 
We have developed a real-time hybrid synthetic aperture detection, imaging and tracking system. This system consists of two camera subgroups, one is used for synthetic aperture detection, as displayed in left column, another one is used for synthetic aperture imaging, as shown in the bottom. Both detection results and imaging results are combined together to make tracking more robust. The greatest advantange of this system is that it can easily solve severe occlusion problem frequently confronted in visual detection and tracking.

Tao Yang, Yanning Zhang, Xiaomin Tong, Xiaoqiang Zhang, Rui Yu. Continuously Tracking and See-through Occlusion Based on A New Hybrid Synthetic Aperture Imaging Model, Proc. of IEEE Conf. on Computer Vision and Pattern Recognition, 2011. [pdf]

Developing a fully automatic, efficient and robust video content analysis system is a subject of great scientific and commercial interest. Intelligent video content analysis with a static camera has been well researched over the past decade, and many excellent algorithms and systems have been proposed in the literature. However, robust video content analysis for moving camera is still a challenge currently, and we saw this technology gap as an opportunity to develop our own advanced video processing algorithms and system,for important applications such as aerial video surveillance, wide-area monitoring, and moving camera based moving object tracking.

The FlyingSowrd was originally developed to perform video stabilization, but recent developments have added new algorithms and greatly improve its effective and efficiency. Currently, FlyingSword is a real-time system capable of performing registration, mosaicing, stabilization, moving object detection,tracking of videos taken from airborne and ground-based moving platforms(e.g. UAVs, aircraft, robot, intelligent vehicle, active surveillance camera).

The FlyingSword System mainly contains two components: (1) Global motion compensation, and(2)moving object detection and tracking. Global motion compensation. Motion compensation is the premise and key technology of aerial video stabilization, panorama stitching and ground moving target detection and tracking. In FlyingSword System, we develop a novel scene complexity and invariant feature based motion video registration algorithm. Detecting moving objects automatically is a key component of an automatic visual surveillance and tracking system. In many application fields such as airborne surveillance, the moving objects (car, people) may be small, sometimes even color information is not available (thermal video). To handle this problem, we use Motion Histogram Image (MHI) and cumulative object motion over an image sub-sequence for foreground segmentation. Tracking is the fundamental block for the high level content analysis and exploitation. Currently, blob tracking is implemented for its simplicity and efficiency, we implement Global Nearest Neighbor (GNN) for data association, and similarity scores between tracks and new measured blobs are estimated by computing their spatial distance. For occlusion handling, we maintain object moving direction, velocity as well as object appearance model. To deal with broken trajectories, a post-processing algorithm is under developed to create a global tracking trajectory.

Automatic counting of passenger is very important for both business and security applications. This project takes a single camera based vision system which is able to count passenger in a highly crowded situation at the entrance of traffic bus. The unique characteristics of the proposed system include: (1) A novel feature point tracking and online clustering based passenger counting framework is presented, which performs much better than those of background modeling and foreground blob tracking based methods. Moreover, this framework is general and can be easily implemented in other passenger counting application fields. (2) A simple and highly accurate clustering algorithm is developed, which projects the high dimensional feature point trajectories into a two dimensional feature space by their appearing and disappearing time, and count the number of people through online clustering. (3) All test video sequences in the experiment are captured from real traffic bus in ShangHai city, and the results show that the system can process two 320x240 video sequences at a frame rate of 25fps simultaneously, and count passengers reliably in various difficult scenarios with complex interaction and occlusion among people, achieves high accuracy rates up to 96.5%.

TaoYang, Yanning Zhang, DapeiShao, YingLi. Clustering method for counting passenger getting in a bus with single camera. Optical Engineering, 49(037203), March 2010 [pdf]

1. TaoYang, Francine Chen, Don Kimber, Jim Vaughan. Robust people detection and tracking in a multi-camera indoor visual surveillance system. IEEE International Conference on Multimedia & Expo 2007(ICME), Beijing, China, 2007, 675~678
2. Andreas Girgensohn, Don Kimber, Jim Vaughan, Tao Yang, Frank Shipman, Thea Turner, Eleanor Rieffel, Lynn Wilcox, Francine Chen, Tony Dunnigan. DOTS: Support for effective video surveillance. ACM Multimedia 2007(ACM_MM, Full Paper), Augsburg, Germany, September 2007, 423~432

We have developed an approach to accelerate 3D reconstruction by introducing a mechanism of vertices sharing during the process of traditional voxel splitting. Our system can run at 10fps with 8 cameras on a PC with configuration: i7-950 CPU, 4G RAM.