Research And Application Of Multi-object Tracking

With rapid development of society, the public safety incidents, especially the violent terrorist incidents, have become increasing, so there are new requirements for traditional video surveillance system. Bec ause the intelligent video surveillance, as the new research results of video surveillance, meets the current developing trend of the information industry and is combined with the computer vision and machine learning, it gets widely attention. There are ce rtain research value and broad application prospects in it. The multi-object tracking is the core component of intelligent video surveillance, so it is extensively studied by researchers. Due to the intrinsic complexity of the mu lti-object tracking field, there are still lots of hard problems to be solved. This paper focuses on intelligent video surveillance in multi-target tracking algorithm and its application. The main work and innovations are summarized as below:1. An object detection algorithm which bas es on region convolutional neural network(CNN) is designed. The CNN is brought into the algorithm to solve variety problems of artificial object feature. Meanwhile the region select algorithm which produces suspected region that may contain objects is prop osed to meet the real time needs of intelligent video surveillance system. A CNN based on the region select algorithm is designed to improve the traditional CNN algorithm. The experiments show that the algorithm extracts higher resolution object feature and has better performance in accuracy of object detection.2. An object tracking algorithm which bases on basic color feature and adaptive scale factor is proposed. The algorithm uses basic color feature and relative position of objects to solve the data asso ciation problem that is brought by object occlusion. The structured support vector machine is used to compute the best configuration and finishes data association. Meanwhile the scale update algorithm which bases on object information changes is used to co rrect objects' position. The contrast experiments show that the algorithm has better performance of accuracy and correctness of object tracking.3. An intelligent video surveillance system of airport is designed and implemented based on the above algorithms. The system consists of server and client. The client uses detection algorithm of region select CNN to detect aircrafts and sends results to server through network. The server receives detection results and uses tracking algorithm which bases on basic colo r feature and adaptive scale factor to track aircrafts and warns when violations happen. The system achieves the goal of real-time intelligent surveillance and can automatic analysis trajectories.