Visual Robot Vision Tracking In Sports Background

Machine vision has a wide range of applications in the modern manufacturing industry,the manufacturing process of moving object detection and tracking has become a hot topic in the field of modern manufacturing.At present,in the production of industrial robots,the general use of teaching or offline programming work,can only repeat the pre-set good action,the processing of objects and the work environment perception is relatively low.It is a new trend in the industrial pipeline that the machine vision technology be applied to robotics in order to achieve automatic classification and assembly task.So,using machine vision algorithm to track the part images and to achieve automation is of great theoretical significance and practical value on industrial production lines.In this paper,the research on visual tracking in moving background based on machine vision industrial robot system will be carried out.Actually,the illumination changes,background clutter,object motion blur,scale change,fast moving of the target and so on.All those factors the scale transformation,rotation and other factors that affect the tracking effect are uncertain.However,there are so many adverse factors in the real world like illumination changes,object motion blur,scale change,fast moving of the target,so part image tracking is still a challenging.In order to improve the accuracy,robustness and real-time of target tracking algorithm,a stable object tracking algorithm is proposed.The algorithm combined tracking with kernelized correlation filters.The specific research works are as follows:Firstly,there is a brief introduction about industrial robot visual system including that the components of industrial robot and analyzing principles of selecting the industrial cameras to complete the experimental platform.Secondly,the basic principle and main implementation steps of the tracking with kernelized correlation filters are introduced in detail.The basic process of continuous tracking target work-piece is analyzed.Then,the experiment proves that the tracking is effective and finds some problems.Thirdly,in order to solve the problem of drift and scale update failure in the process of object tracking,the original kernel correlation filters is improved from two aspects.When using the ridge regression model to train the classifier,we consider the temporal and spatial relations of two consecutive frames,and join the classifier to enhance the stability of the classifier,and avoid the tracking drift.Besides,scale estimation is introduced after the acquisition of the object position by the tracking kernel correlation filters.We add the color histogram model of the target in the Mean Shift algorithm to get the weight distribution map of the candidate region,and then calculate the size of the target scale according to the first-order statistical feature of the weight distribution graph.Finally,the algorithm of tracking is used to test the video sequence of the moving parts collected by the camera.The experimental results show that the improved tracking algorithm can adapt well to the target scale,different brightness,fast motion,background blur and so on.While the average center position error,the average success rate,the speed have a good advantage.Therefore,the improved tracking kernel correlation filters have strong robustness and tracking efficiency for the target image tracking on the automatic production line.