The Research On Object Detection And Tracking Algorithm Based On The Video

With the rapid development of computer science and mediatechnology, the computer vision technology has become one of greatconcern in various fields. As one of the most challenging research focusin the field of computer vision, moving object detection and trackingbased on video stream relate to multiple disciplines, such as video imageprocessing, pattern recognition, artificial intelligence and so on. It hasbeen widely used in industrial, military, medical, remote sensing andmany other fields. Therefore, the research of moving object detection andtracking has great theoretical and practical value. In this paper, the studywork has been done with the relate algorithm involved in the process ofmoving object detection and tracking and make improvements about thealgorithm, the main work is as follows:In object detection, from the algorithm of the concepts, principles,processes and key technologies, we have had a detailed analyze of themost common moving object detection algorithm. Validation andcomparison by experiment, then analyzes and summarizes the advantagesand disadvantages of each algorithm. In order to solve the puzzles in theprocessing of the Gaussian mixture model parameters estimating andmodel selecting, the research has been done as follows: An improved EM(Expectation-maximization) algorithm has been proposed using themaximum punishment of the likelihood function because of the defectthat EM is easy to fall into local optimum of the solution space when weestimate parameters of GMM (Gaussian Mixture Model); When the lightchange, the model will fail. For this Problem, we propose a new failuredetection method. And we use three frame differences instead of GMM.For the slow moving objects into the background and the stationary beganto move. I proposed a new parameter update method in this paper, until itconverges and then re-use of this model, to avoid the influence broughtabout by the large amount of noise due to light mutation.In object tracking, for the problem of nuclear window widthunchanged during the object tracking, the edge detection algorithms extract the target measure is introduced to self-adapt the Mean Shiftalgorithm of changing in target scale issues. Meanwhile, For the MeanShift algorithm can not meet the goal of rapid and large area of velocityshelter problem, the Kalman algorithm is introduced to predict the targetlocation, and then combine the improved Mean Shift algorithm search thetrue location. The method of the paper has effectively improved theaccuracy and robustness of the tracking algorithm.