| Summary: | 碩士 === 國立宜蘭大學 === 資訊工程研究所碩士班 === 97 === Multiple cameras tracking system is an important research topic due to the rapid development of computer vision and the increasing popularity of monitoring system devices. In this thesis, we will propose a real-time smart multiple cameras people tracking surveillance system. The tracking system can automatically track the moving people of different camera views. The system also enables users to define the meaning of optimal view in order to meet different application requirements.
The first step of our proposed multiple cameras people tracking system is foreground detection. We use a Simplified Gaussian Model for background modeling quickly and detecting foreground object effectively. For handling occluded objects, an occlusion detection function based on an M-to-One (Multiple Points to One Region) relationship with only one homography matching can achieve high computational efficiency. Using the homography matrix technique, object point of different cameras can project on the established top-view. Through the distances between the intersection of principal axis and object point of top-view, the correspondence of objects in different camera views can be solved. Finally, we used Kalman filter to integrate and track object position. For the selection of optimal view in different cameras, a technique based on WOVP (Weight Optimal View Probability) approach is proposed to automatically select optimal view that user defined for multiple cameras people tracking system.
|
|---|
