Research On Apparent Model-based Visual Tracking And System Implementation

Visual tracking which combines a large number of advanced technologies and research production of disciplines of pattern recognition, artificial intelligence, and digital image is an important field and research focus of computer vision research directions and has much real value in social field. Currently, visual tracking technology has been used in security surveillance, traffic control, intelligent robots, video games as well as military and other fields. However, it is still a difficult problem to make it more robust, faster and continues long to track flexible target whose appearance change significantly. This paper conducts in-depth research on online learning tracking and improved both the effect of flexible target tracking and the robustness and accuracy of visual tracking by optimizing the rules of online learning tracking algorithm and introducing adaptive target color vector in target description.Firstly, this paper gives a brief introduction of target modeling and status at home and abroad of visual tracking. This paper analyzes the common characteristics as well as the advantages and disadvantages of target modeling. Combining the pedestrian target unique outer color distribution characteristics, this paper proposes an adaptive block color vector characteristics, and describes the feature extraction methods in detail. The feature in the description of the pedestrian target is simple and it has a high match speed as well as a low requirement of video resolution. Subsequently, the adaptive block color vector features with a sliding window of detection methods were combined to verify the effectiveness and timeliness of the characteristics of target detection process.Secondly, this paper analyzes the online learning tracking method and introduces a method based on combination of detection and track. The method fuses the detection results and tracking results to obtain the changing appearance model of target in the process of campaign and put the appearance model into the model database. It can solve the problem of appearance changing in the process of target tracking. Also this paper summarizes the constraints of detection results and tracking results to enhance the online learning tracking method in practical applications.Finally, an online learning tracking experimental platform is established based on exist equipment in library. During the implementation of the platform, modules are chosen carefully to get rid of trouble when running software on different computers, for example, CSerialPort class is selected instead of MSComm in serial communication, and SQLite is chosen to replace other SQL Server database. At the same time, the platform has reserved a number of interfaces to facilitate post-algorithm modifications. At last, an actual operation results is given, which shows that the system is able to achieve the expected goal commendably.