Heterogeneous Camera Network For Object Tracking

Object tracking is hot topic of computer vision, has a profound theory value and broad application prospect. Heterogeneous camera network composed of PTZ cameras and omni-camera can realize coverage of large scale area and capturing object detail images, and is widely applied in video surveillance and object tracking.This thesis focuses on heterogeneous camera network composed of PTZ cameras and omni-camera, and gives more attention to increasing robustness and accuracy of object tracking and effectively capturing valid object detail images. The main contribution can be described as follows:First, existing heterogeneous camera networks have low tracking accuracy and low robustness. A cooperation tracking algorithm based on PTZ camera and omni-camera is proposed to solve this problem, which is composed of data association based on multi-camera measurement, data fusion based on measurement and object tracking based on Kalman filter.Second, aiming at the demand of capturing object detail images, an optimized multi-PTZ cameras cooperation algorithm is proposed based on VDOA (Visibility, Distance and Orientation, Angle) to select the most suitable PTZ camera to capture facial image of objects. The ability of capturing object detail images is increased.Finally,single object tracking and multi-objects tracking experiments are carried out using an experiment heterogeneous network composed of one omni-camera and two PTZ cameras. Experiments results verify the multi-sensor information fusion algorithm and multi-camera cooperation algorithm.