Adaptive Strategy Based-Time Object Tracking

The traditional video monitoring system only provides the capture, storage and play back and other simple functions of the video, it is difficult to play the role of early warning and alarm. However, the intelligent video monitoring system use computer vision technology of video signal processing, analysis and understanding to allow the computer to filter the goal of information which the user does not care, so as to provide helpful information in monitoring for the user. Under the condition of unmanned monitoring, It can automatically identify and track through the analysis of the changes in the monitoring scene. And it can judge and understand the behavior of the target on the basis and alarm in the first time or provide helpful information to the control room when the abnormal behavior happened.The target tracking of video images is a hot issue in computer vision technology.Target tracking is estimating the correct position of the target in video sequence which plays an important role in human-computer, behavior recognition, motion analysis, video surveillance, robot technology and other application. Because of the complex interferential factors in the natural scene,such as overall occlusion, partial occlusion, plane rotation, motion blur, illumination changes, background of chaos and the change of scale, the existing tracking algorithm still exist deficiencies in speed and accuracy.In order to improve the accuracy and robustness of the video target tracking in complex environment. Firstly, this paper analyzes the deficiency of target tracking algorithm based on color attribute, and through improving the method of the calculate the current frame of the tracking sample when current frame and the previous frame difference is small, considering the current frame and the previous frame, to get the best training sample by giving different weights. On the contrary, we will get the best training sample through the current frame calculation when the difference is large. Thus it would make the algorithm more robust and achieve real-time speed by all of the above methods.Occlusion is one of the problems which the video tracking processing often encounter.Because the method color attribute tracking algorithm search the target area is similar to template matching whether we synthesize the data of previous frame and current frame to be the training sample or only use the data of the current frame or previous frame to be the training sample, they all are difficult to obtain good tracking effect. In this case, this paper try to use the sparse collaborative appearance model algorithm which is more robust but its computational complexity is more difficult to recalculate the location of the current frame. The deficiencies of the method is its high computational complexity, it would be difficult achieve the real-time speed after long time.In order to achieve fast and accurate tacking, an adaptive strategy selection mechanism is proposed to integrate the two algorithms through confidence evaluation of the tracking results, and put forward a kind of adaptive strategy based real-time object tracking. Experimental results on multiple public datasets show that, compared with the existing object tracking algorithms, the proposed method is accurate and fast. It performs well in regard of serious occlusion, illumination variation and motion blur.