Research On Target Tracking Algorithm Based On Sparse And Compressed Perception

With the rapid development of computer, people more hope that the computer can complete human visual information processing. And then produced a new discipline, namely, computer vision. Target tracking technology belongs to a branch of computer vision. Target tracking technology can be used in security monitoring, intelligent transportation and other fields. This paper mainly studies the target object tracking algorithm in the image sequence.Online tracking is a challenging problem. This challenging mainly comes from when caused by internal cause and external cause change an object’s surface, and the need to establish an effective external representation model of object. The internal cause and external cause of the change of the object surface respectively is: the diversity of the moving target posture, the change of background light around and background of the object shelter and movement caused by the fuzzy etc. In view of the above factors, the effective tracking model is of great significance. Tracking algorithm model mainly includes motion model and observation model. Motion model of target tracking algorithm mainly complete the prediction of the moving objects, among them, the prediction is based on a frame or a few frames before the image. Measurement model mainly includes the external description of objects, matching, and updated. The external description of objects required to completely describe the target object characteristics, and to reduce the time of the track calculation, at the same time can be used to update the model in the process of tracking.In this paper, online sparse model of target tracking. First of all, this model using principal component analysis to extract the target object is the main feature of the image, and the current frame image is expressed as the sum of the main features and the secondary form. In order to achieve the purpose of tracking, the sparse representation of the target object online with new methods. Drift phenomena appeared in the second, in order to avoid tracking, the proposed model takes into account because of the background object occlusion and the target object motion blur caused by the drift phenomenon, rather than simply contain image observation model updates. Finally, this paper through the experiment and for evaluation of the algorithm in this paper, the algorithm is verified and evaluated. The experiment part includes: in this paper experiment and the light changes. Through comparing several experiments, the stability of the algorithm in this paper shows the stronger.In addition, this paper designed a kind of efficient tracking algorithm, this algorithm is observed in the model feature extraction in compressed domain perception. In the compressed domain, using one of the main features of the extracted from the image, so as to establish the measurement model can very well said to track targets. Target tracking algorithm based on compression perception, realization process is as follows: according to sparse perception theory, through a RIP conditions very sparse matrix measurement characteristics of original image space projection, you can get a low dimensional compression subspace. Low dimensional compression subspace can be very good keep the characteristics of high dimensional image space information. So we through sparse measurement matrix to extract foreground the characteristics of the target and the background, as updated online learning classifier of positive samples and negative samples, and then use the naive Bayesian classifier to classify the next frame image target image slices under test.Finally, the text in the evaluation of algorithm introduced two evaluation criteria: overlap rate(overlap rate) to assess the margin of Error and the Center position(Center Error). With the other algorithms in the center of the overlapping rate and the location of the error rate, this algorithm has higher the center of the overlapping rate and low error rate. In this paper, the experimental results show that the algorithm in the realization of movement target tracking with strong robustness.