2A1-D10 Segmentation and Correspondence of Human Body from Multiple Camera System
Hyun-Uk Chae, Taeho Kim, Kang-Hyun Jo · The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec) · 2009
This paper describes a method to make correspondence people across multiple-views. For obtaining correspondences, the system has three steps such as motion detection, color segmentation and correspondence of people. The moving person is detected by subtraction from multiple background model. We in a process generate multiple background model from image sequence to obtain human silhouette. Furthermore, each detected silhouette from different views is segmented by specific criterion which consists of color and context information. The segmented region is represented as blobs with color distribution and a human consists of several blobs. We then estimate Gaussian mixture model from the set of blobs. Finally each region of blobs is matched with other blobs from different views to identify a same person.