Development Of An Autonomous Target Tracking System Based On Machine Vision

Teleoperation robot is a kind of remote control robot,which can carry out the behavior instruction sent by the operator remotely.It is because of the remote characteristics of teleoperation robot that is often used in submarine exploration,aerospace engineering,telemedicine and other fields.Especially in special or strange environment,teleoperation robot is widely used.However,in an unfamiliar environment,people have limited understanding of the working environment of the teleoperator.During the operation of the robot,due to the limited reconnaissance distance of the navigation equipment they carry,it is difficult to timely detect the bad road conditions ahead,which makes it more difficult for operators to control the teleoperator.Therefore,this paper develops an autonomous tracking system based on machine vision to monitor and track the teleoperation robot from the air.In this way' the position of the teleoperation robot can be located,and the surrounding environment of the robot can be fed back to the operator in time,so as to give appropriate instructions to the robot.But using the aircraft equipped with visual tracking system to track the target can overcome the disadvantages.However,visual tracking depends on the intensity of light on the target.This paper develops an autonomous tracking system which can adapt to the change of light intensity from strong to weak and regard the teleoperator as the tracking target.First,an algorithm called Haar cascade classifier which based on the AdaBoost,are selected in this theory.After studying the principles,a database is founded.And the database contains positive and negative sample of the earth-moon teleoperator,by using learning training method,a cascade classifier obtained,and the classifier can identify the teleoperator effectively.After that,choosing a good tracking algorithm as a key to the paper.In this part the theory of Meanshift and Camshift are learned in-depth.Comparing the two algorithm by making experiment.Camshift is better than Meanshift and selected as a tracking method in this paper.The following part is the most important thing in the whole thesis.This part describe how to achieve the auto-detect and tracking system.Through experiments,it is verified that this method can autonomously track the target in sufficient light and dim light under the condition of no shielding.On the basis of target detection and Camshift tracking the teleoperator,combining the two algorithm can achieve the work.But as the Camshift cannot cope with occlusion and color interference Kalman filter has been used to improve the tracking quality Never the less,color interference is still worse.In this case,the combining Gaussian background modeling into the algorithms,and by using this method,background and foreground are separated easily,and the color interference has been disappeared and then the quality of tracking improved.At last,Summarize the whole thesis,at the same time come up with some future work which need improved in small target tracking and fast moving target tracking.