Research On Target Tracking Technology Based On Machine Learning

With the development of science and technology,machine vision has become one of the main ways for humans to perceive the world and obtain important information.Target tracking is an important research direction of machine vision,and is widely used in intelligent video surveillance,medical image review,human-machine Interactive,all-day climate reconnaissance and military applications.In the actual tracking process,due to the complexity,particularity of the application environment and the type diversity of the tracked targets,the target detection and tracking algorithms are poorly robust,difficult to track for a long time,poor stability,poor real-time performance,and low tracking accuracy.In view of the above problems,this paper has carried out research on target tracking technology based on machine learning.The main research contents are as follows:(1)A tracking algorithm based on Camshift and Kalman filtering.Aiming at the problem that Meanshift has poor tracking effect and no tracking of tracking frame,the tracking algorithm combined with Camshift and Kalman filtering is studied.In order to verify the anti-jamming effect of the algorithm on the color,the short video was tracked and analyzed.Experiments show that the algorithm combined with Camshift and Kalman filtering has better tracking effect,and is not interfered by background color,and there is no tracking failure phenomenon.(2)The theoretical knowledge,basic principles and algorithm characteristics of the TLD tracking algorithm are studied,and the advantages and disadvantages of the algorithm are analyzed through experiments.In the template matching experiment process of this paper,when comparing the relationship between positive and negative samples and time-consuming,it is found that the original TLD tracking algorithm takes a long time.However,in the tracking of the target pedestrian occlusion,the tracking performance of the TLD algorithm is high when the target scale,shape change,and partial occlusion and the occlusion time are short.(3)Optimize the original TLD algorithm.For the original TLD algorithm,the amount of calculation is large.When there are similar targets and the target is occluded seriously,the tracking accuracy is low and the effect is poor.This paper proposes an optimization algorithm—using the Camshift algorithm instead of the light in the original TLD algorithm tracker.Flow method.By comparing the tracking experiments of pedestrians,it is found that the optimized TLD algorithm has no hysteresis,and can accurately track the moving targets with more occlusion and slower motion.(4)TLD tracking algorithm combined with Camshift and Kalman filtering.In the long-term tracking of the target,for the problems of occlusion,fast motion,similar target influence,illumination change,etc.,this paper uses TLD tracking algorithm combined with Camshift and Kalman filtering.Experiments on David video tracking show that the algorithm has higher success rate than the original TLD algorithm and good tracking effect.At the same time,the video frame center position error is the smallest,concentrated within 20 pixels.The real-time and rapidity of tracking in the experiment process is good.In order to verify the tracking robustness and real-time performance of the proposed algorithm,this paper uses the tracking of athletes to verify the experiment.The experimental results show that the overlap degree of the algorithm is stable,and it is robust to target tracking with fast running speed,similar target interference,serious scale change and occlusion.The center offset distance is low,which indicates that the algorithm has good tracking effect and good real-time performance.