Research On Binocular Vision Obstacle Avoidance Of Indoor Mobile Robot Based On ARM

With the popularization of service robots and sweeping robots,mobile robot research has currently become one of the directions of robots.The research on the environment perception ability and path planning of mobile robots has brought it into a more intelligent times.As sensing sensor of the robot,the binocular camera has the characteristics of strong operability and more intelligent processing.Especially in recent years,the extensive use of deep learning in the field of computer vision has made the application of cameras as the sensor of robot platforms more and more common.Therefore,this topic of this thesis selects research on binocular vision obstacle avoidance of indoor mobile robot based on ARM.The main researches in this thesis include the basic principles of binocular vision and camera distortion correction,binocular camera stereo matching,image recognition and object detection,and robot path planning.First of all,This thesis analyzed the basic principles of binocular vision,the causes of camera distortion,and the principles and methods of camera calibration.And then,This thesis derived the change matrix of the camera from the image pixel coordinate system to the world coordinate system,used the Zhengyou Zhang’ s checkerboard calibration method to perform binocular camera calibration.then used the camera’s relevant parameters which is obtained from calibration to correct the image and restored 3D information.Secondly,this thesis proposed an idea based on image pyramid for the shortcomings of stereo matching on HD images.Similar to the human observing the world,people often do not need to spend a lot of energy on nearby objects,but when observing distant objects,they need to observe them carefully to obtain detailed information.Through this stereo matching method,the accuracy can be satisfied and the speed of stereo matching can be ensured.Thirdly,the YOLOv3 algorithm has been used the in this thesis for detection.The traditional image features such as HOG or SIFT has slow speed of detection and the terrible result.Using deep learning to detect can ensure accurate object detection.Applying GPU acceleration ensures real-time and accurate object detection.The artificial potential field method is improved to planing path for the robot in this thesis.The artificial potential field method is simple,and it is easy to improve the algorithm for different application scenarios.However,the artificial potential field is to simulate the effect of the potential field,so there will be several disadvantages such as local minimum and the inability of the robot to reach the target point.Improvements are made to the above problems to ensure that the robot can accurately complete the path planning.Finally,an indoor mobile robot platform based on ARM has been designed for experiments in this thesis.The advantages of the improved stereo matching algorithm in matching speed and the feasibility of the improved artificial potential field method in path planning are verified by experiments.