Research On The Robot Obstacle Avoidance And Target Tracking Based On Computer Vision

Autonomous navigation is an important symbol of mobile robot to achieve intelligent,which is the key factor to endow the robot with autonomous perception and action ability.Obstacle Avoidance is a basic requirement for autonomous navigation of mobile robot,therefore,it is very important to study the obstacle avoidance of robot in the case of unknown environment.At the same time,with the requirement of robot's practical application,the target detection and tracking technology of mobile robot has been more and more widely studied.This paper uses Turtlebot as the research and implementation platform of mobile robots,using monocular and binocular cameras as the main sensors,studied the problem of robot obstacle avoidance in unknown environment and the problem of moving objects detection and tracking problems,put forward the corresponding algorithm and implementation plan.The main research work and contributions of this paper include:(1)In view of the passive obstacle avoidance problem of robot in complex and unknown environment,this paper designs a novel obstacle avoidance system based on monocular vision.The system uses the feature scale detector to detect obstacle in the front of the visual field of robot,and uses the optical flow balance strategy to avoid the obstacle on both sides,for obstacle with low texture,the image entropy method is applied.(2)A Marker tracking method based on monocular camera is studied.The detection and recognition of the markers are carried out,then the matching relation between the 3D points in the space and the 2D feature points in the image plane is extracted,next the pose of the robot relative to the Marker is calculated.Finally the pose is used as the input of motion control,and the speed control to the robot is carried out to realize the tracking of the given Marker target.(3)For any ordinary target detection and tracking,first of all,based on the convolution neural network,detecting object in the image;after detection to the object,then based on correlation filter to track the target in the video;finally,using binocular camera to carry out sparse 3D reconstruction,then pose of the robot relative to the target can be restored,the poseinformation is used to control the robot to follow the target.In this paper,based on the depth information,the target scale ambiguity problem of the correlation filter is improved.