Target Tracking Of Small Mobile Robot Guided By Motion Model

With the development of robot technology,the potential demand for service robots is increasing.Pedestrian following robot is a kind of service robots.In order to complete the task of robot following the target,firstly,a robust visual tracking algorithm is needed to track the target,secondly,a stable robot motion control algorithm is needed.In this paper,the characteristics of human target tracking small mobile robots is analyzed Based on the real-time kernel correlation filter(KCF)visual tracking algorithm,a kernel correlation filter guided by motion model(MM-KCF)is proposed to fulfill the human feet robust tracking under the conditions of light change and fast walking.For the occlusion problem in tracking,the kernel correlation filtering algorithm is improved by using adaptive output responses,at the same time,the occlusion detection method based on correlation rate is combined to improve the accuracy and robustness of human feet tracking algorithm under occlusion conditions.Finally,the feet tracking algorithm are verified by using the turnlebot robot in the ROS system.The specific research contents are as follows:1.By studying the characteristics of human feet walking,human feet motion model is established,then a kernel correlation filtering algorithm guided by the motion model is proposed.During tracking,the local and global motion models are established to predict the position of feet in the next frame,and then the target detection region of KCF algorithm is obtained from the predicted the position information.Thus,the tracking precision is improved and the tracking error is reduced.2.By improving the detection region shape of KCF algorithm,the switching problem of target tracking box caused by the high similarity of the feet is solved without affecting the tracking precision.Aiming at the problems of inaccurate detection of target position and unreliable cyclic shift samples,the adaptive response kernel correlation filter guided by motion model(MM-AR-KCF)is proposed.Aiming at the occlusion problem in tracking,the anti occlusion(AO)detection method based on peak and neighborhood correlation is studied,and MM-AR-AO-KCF algorithm is proposed.3.In the actual scene,we use the mobile robot turtlebot to carry out the robot experiment.By improving the adaptive linear speed and angular speed control algorithm,the relative distance between the robot and the target person is kept in a constant range.Experiments are carried out under fast motion,short-time occlusion and long-time occlusion cases respectively.The results show that the average precision of MM-KCF algorithm is 0.769,which is much higher than that of KCF algorithm,BACF(background aware correlation filters)algorithm and SAMF(scale adaptive kernel correlation filters with multiple features)algorithm under the conditions of illumination variation,background clutters and fast motion.The average precision of MM-AR-AO-KCF algorithm is 0.818,which successfully solves the occlusion problem in feet tracking.Experiments show that the algorithm proposed in this paper has strong reliability and real-time performance when it is applied to the human feet tracking of the turtlebot.It can also effectively solve the problem of target loss caused by the occlusion of both feet in walking,and good anti occlusion target tracking is achieved.