Video Tracking Method Based On Spatio-Temporal Context

The field of computer vision is one of the most important part of video target tracking, now video target tracking has been widely used in national defense, science and technology, aviation and national life, which is representative of the application of intelligent video surveillance, intelligent transportation, and human-computer interaction. In view of this, video target tracking is still has great prospects for development.Target tracking process mainly includes target detection, target extraction and target recognition and target tracking, and track the target state is obtained. This state is mainly including its trajectory, such as position, velocity and acceleration information. From this forecast the next frame of tracking information and tracking, finally reach the target tracking behavior understanding.This paper first introduces the research background and research significance of video target tracking, and then to the currently research status at home and abroad in this field has carried on the detailed narration, and the classic in the field of video target tracking algorithm and key technology are summarized and summed up. Through the research and analysis of various target tracking algorithm, this paper presents a new algorithm of target tracking based on segmentation. The algorithm is in middle level visual features, and establish a model of spatial and temporal context based on segmentation, and determining the target position problem as a goal in the probability of a certain location problem, namely the target has the highest probability location for the new target location in a frame.Then with the introduction of the TLD target tracking algorithm in detail, the TLD target tracking algorithm has the advantage of a good learning ability to restore and certain ability to resist shade, but when the target rotating or cover more than 50%, tracking failure would happen. In this paper, the target tracker of the TLD framework has been improved, the context of time and space tracking algorithm based on segmentation on the improvement of the improved tracker can avoid to a certain extent by the target rotation, light intensity change, similar background interference and tracking brought by the event of failure.Finally in this paper, through a large number of experiments show that this improved algorithm can solve the visual tracking field at the same time facing the rotation, occlusions, illumination changes and so on the many kinds of problems, and the algorithm is more adaptive and effective than ever before.