Object Detection And Tracking In Video Image

Moving target detection and tracking is one of the most important subjects in image coding technology and computer vision. It has very important practical value in video surveillance, visual navigation, intelligent transportation, video image compression and transmission.Object detection is at the bottom of the video sequences processing, which will directly affect the accuracy of following advanced applications. This paper introduces the current mainstream object detection methods including, intra-frame subtraction and Gauss background updating. Combing with morphology filter , the object can be extracted from pictures.In computer vision, tracking refers to the task of generating the trajectories of the moving objects by computing its motion in a sequence of images. Numerous approaches have been dedicated to computing the translation of an object in consecutive frames, among which the MeanShift method is one of the most common methods which is also used in the commercial applications. MeanShift is a nonparametric density estimator which iteratively computes the nearest mode of a sample distribution. In most situations it can guarantee the accuracy and real-time tracking, which is a kind of fast and effective tracking algorithm.However, MeanShift algorithm doesn't use the target's motion direction and speed information in process of target tracking. So it brings about failures in fast motion target tracking. An algorithm combined center of gravity with MeanShift algorithm is proposed in this paper. At first, we can use the centroid as initial position; and then MeanShift iteration is done in the location of the centroid; and the bhattacharyya's coefficient is applied to judge the matching degree between the current target and reference target. Experimental results show that the new algorithm can help achieve fast and effective object tracking.In addition, when the target is totally occluded by obstacles, MeanShift algorithm mistakes obstacles for a possible target model. If the follow-up frame occur in the moving targets, it will not be effective . To solve the problem, this paper adopts MeanShift algorithm based on kalman filter. According to former movement targets information, kalman filter can estimate the initial position at the next moment, and then MeanShift algorithm can iterate it based on kalman filter. Compared with the improved algorithm and the traditional MeanShift, the improved method can obtain a stable tracking result when the target is totally occluded. The effect of the improved algorithm has been significantly improved.