Research On Targets Tracking Algorithm Based On Spatio-temporal Context Model

With the development of high performance computers,low cost video cameras of high quality become widely used,such that the demand in automated video analysis is greatly increasing.This leads to wide applications of object tracking in automated surveillance,intelligent transportation system,human computer interaction and vehicle navigation.Although researchers have proposed a variety of vision based target tracking algorithms for tracking problems of different types of environment and different types of targets,however,it is still facing many difficulties in practical application of video object tracking,such as background clutter,illumination changes,partial or full occlusion,the changes in target appearance,fast moving of the target and so on.All those factors will lead to a result mistake in the target tracking.Therefore improving the target tracking algorithm and improving the accuracy,robustness and real-time of target tracking algorithm has always been a hot and main difficulty in target tracking research system.This paper had studied the classic algorithm of the video target tracking first,and then proposed a target tracking algorithm under complex environment.The algorithm combined temporal and spatial contextual information as the main idea.The main innovation points of this paper are as follows:(1)Aiming at the tracking drift problem due to object model update falsely in the online learning tracking algorithms,a simple but efficient solution is proposed.In the target area point trackers are uniformly sampled,which are assessed based on texture description in two consecutive frames point trackers and then the initial location of target is completed.Multi-dimensional feature spatio-temporal context model is used to output the precise position of object by the confidence map,the model update rate is decided combining with the confidence map and a multi-scale update mechanism is proposed.(2)Proposing 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 in the next frame.