Design And Implementation Of The Compressed Domain Video Moving Object Tracking System Based On HEVC

Video moving object tracking is one of the difficult problems in computer vision and a critical, time-consuming step in video processing, the research and implementation of a high efficiency object tracking has significant value. The main stream procedure for current object tracking systems as follows: 1) Decode the video to restore the texture information for every frame. 2) Extract image features and motion information from the restored image, and implement video object tracking based on these features and information. Notice that the compressed format is another completed representation of video, and massive moving information has been exploited during the video encoding process. In this paper we will discuss whether these compressed-domain information can be utilized to implement video object tracking.To cope with the aforementioned problems, we discuss video object tracking in the compressed-domain, and we also designed and implemented a compress-domain moving object tracking system using the information extracted from the HEVC(High Efficiency Video Coding) compliant video stream. The significance of the research can be concluded as followings: Firstly, it implements object tracking in the compress-domain, which reduce the computational complexity by eliminating the texture information decoding and image feature extraction process. Secondly, the system has successfully realized the tracking algorithm on HEVC, HEVC is the state-of-the-art video coding standard nowadays, it is said to achieved great efficiency improved compared to the previous standards and thus very promising. Moreover, our system provides data analysis and visualization, making it convenient for further analytic and research purposes.The primary work of this paper include:(1) Design a component to retrieve the compressed-domain information from video bit streams.(2) design and implementation of the video moving object tracking framework, and built a HEVC tracking system on top of this framework. The framework also provides mechanisms to allow user to extend and customize as need.(3) Design and implementation of a User Interface that can be used to analyze tracking result and calibrating the system parameters.By now the development of the prototype object tracking system has already been finished. The test results on some of the standard test sequences show that the system has an average accuracy about 70% and the average recall is about 75%.