Target Tracking Prediction Research Based On Computer Vision

In modern high-speed development of technology, computer vision has become a comprehensive discipline that gain more and more attention.By means of computer vision, tracking and forecasting of moving objects has become an essential part of anational defense and military technology, industrial production and modern agriculture.Under the topic of computer vision, target tracking is a multi-disciplinary subject which is blend of image processing, intelligent systems and process control.In computer vision applications has superior advanced nature.In applications all over every aspect of life,especially,based on computer vision of the moving object tracking and forecast in the future war has important significance.It is widely used in the usual way, the vehicle bank atms, parking lot, residential area, etc. in order to prevent the occurrence of theft and robbery, also widely applied in the protection of social and public safety.Target tracking and forecasting of the target area can use region-based tracking algorithm. While background finite difference method is one of the most basic purpose method to achieve tracking and forecasting purposes.Background subtraction method is collecting the background of the scene image, the background image frame and the frames of the current to do XOR or subtraction operation, in order to achieve the target recognition and tracking purposes.In the actual track, due to camera shake, the changes in light conditions increased target tracking and other reasons, and target tracking may appear coincident, occlusion, contour changes, among other factors, making the tracking and forecasting of reliability and accuracy face great challenge.Most of the existing track prediction algorithm discusse the target tracking, but ignore the forecast of the target’s next move.Therefore, based on computer vision, computer vision software halcon10.0, background estimation and Markov chain model,This paper do in-depth study on a moving object multi-objective track prediction, proposing background estimation tracking based halcon10.0, and the Markov chain model applied to the image of the multi-objective track forecast; and from the experimental results and theoretical studies the two parties verify the reliability and efficiency of the proposed theory.The main content of this paper is as follows:(1) Introduces the current situation of the research target tracking based on computer vision, and describes the existing based on computer vision video tracking technology, analyzes the target tracking system based on computer vision, and videos tracking prediction of the corresponding technology in this paper.(2) Based on halcon10.0background estimation method of tracking, this paper used halcon10.0software to realize to the target tracking detection, firstly, halcon10.0software and its applications are described;Use halcon10.0to read into the video, the first to use halcon10.0call the relevant function on the video image sequence of basic processing, image processing is directly related to target tracking accuracy and reliability. In this paper the basic processing of the image including thresholding, obtain the target area and remove noise. Then use halcon10.0for video background modeling of the background image, and use to realize the background image identification and tracking of the current frame image.(3) For the image segmentation in target tracking,the agglomerative clustering for image segmentation or image data type of semi-supervised is put forward and it can dynamically determine the areas which are divided by image,that is the number of clusters.In this paper, the method to determine the number of clustering in the existing semi-supervised competitive agglomeration clustering algorithm is improved, using genetic algorithms to optimize the search of the cluster center, to improve the accuracy of the algorithm and to reduce the volatility of the number of clusters. According to the nonlinear data situation, the improved algorithm introduce the nuclear distance metric. Nuclear distance is not only good solution to deal with the problem of the nonlinear image but also very effective in the clustering effect of linear data.(4) Forecast based on the Markov chain model described the principle of the Markov chain, the image sub-block on block basis according to the rules set by the specific scene, the transition probability matrix derived according to the rules proposed Markov chain applications to target tracking forecast, the current frame image extracted area of interest (that is the target area) to calculate the transition probability matrix of the Markov chain model to predict the next move of the interest area.(5) The experimental results analysis the tracking more than one vehicle in the road, and do the contrast experiment in accordance with sub-block size; and make the corresponding algorithm with the binomial prediction algorithm and the prediction of the Kalman filter algorithm.According to statistics position deviation to calculate the deviation of the location of the various algorithms, compare them and obtain experimental conclusion.Experimental validation and experimental results prove that:(1)Using the Halcon10.0is to better reduce the processing time of track the target, and by compared with other software processing platform it shows the more superiority of its real-time processing.(2)Using the background estimation method, the quality requirements of the image is not high, the algorithm is simple, and can reduce the target recognition error, can track target precisely without sunscreen.(3)If make the Markov chain model apply to target tracking forecast, it can reduce the scope of the search when predict the action of the target, make forecasting more accurate, faster, greatly improving the efficiency of target tracking forecast.