Detection And Tracking Multiple Moving Objects Based On Omnidirectional Vision

Multiple moving objects detection and tracking relates to image processing and pattern recognition areas. It is an important issue in computer vision. A novel omnidirectional vision sensor which captures real-time 360 degree scene is put forward in this paper. A new multiple objects detection and tracking method is brought up and experimented. On the basis of summarizing current research of multiple objects tracking both at home and abroad nowadays, the main content of this paper is as follows:In respect of moving target detection, a method of moving objects detection based on frame difference and background subtraction is brought up. Images of movement areas are binarized, whose noises are eliminated by morphological method. And all complete moving objects are extracted through labeled connecting area and fragments combination, whose information of all physical parameters is provided. Because of updating background real-time, precision of target detection is improved.An algorithm based on Kalman filter and matching matrix is designed for multiple objects tracking, which uses Kalman filter to get objects status prediction. In multiple objects tracking, complicated situations such as multiple objects shelter, shelter targets separation, target disappearance and new target appearance are solved by matching matrix. Experiments show that the method tracks several objects efficiently.Finally, an experimental system for multiple objects detection and tracking is developed based on Derectshow technique, OpenCV image processing library and C++ programming, which is proved by experiments.