Research On Obstacle Avoidance Method Of Intelligent Mobile Robot Based On Machine Vision

The dynamic obstacle avoidance technology of mobile robots is a research hotspot in the field of robots,which is also a key technology that affects the reachability and operational capability of mobile robots.Traditional methods for obstacle avoidance of mobile robots are mostly based on the "detection-bypass obstacles" strategy,which is to implement path planning after obstructions are detected to avoid obstacles.These methods can perform well in open space.However,when the space for avoiding obstacles is not wide enough,especially in scenes where the positions and postures of obstacles such as workshops and interiors are changing,the problem of narrow local space and even the only channel is blocked.In this case,the traditional obstacle avoidance method cannot be handled,which in turn affects the operational efficiency of mobile robots.In this thesis,we proposes a method for obstacle avoidance based on binocular range finding and obstacle mobility identification,which can improve the defect of traditional obstacle avoidance methods.The method includes three stages:bypassing obstacles,pushing obstacles,and stopping in situ,effectively expanding the reach of mobile robots increases safety.In addition,by analysing the key technologies,this thesis focuses on the three issues of binocular range finding,obstacle identification,intelligent obstacle avoidance strategy formulation and implementation.In the binocular range measurement,a high-precision camera calibration image acquisition scheme is designed to achieve better calibration accuracy for ordinary USB cameras.The disparity map verification method based on foreground extraction is designed to reduce the mis-match phenomenon in a complex background.Using the SGM algorithm to obtain accurate disparity maps and measuring depth information of obstacles.In the obstacle identification,an obstacle and its mobility identification method based on the target detection technology is proposed.Classifying the common obstacle images in the running scene of the mobile robot,add the labels to the positions and categories of the obstacles,and establish a common scene data set for the mobile robot.The dataset was used to fine-tune the YOLOv2 model,and various obstacles were identified for experiments to achieve high recognition accuracy.In the formulation and implementation of intelligent obstacle avoidance strategies,the design of the method for estimating obstacle avoidance space is combined with the identification of obstacle mobility to design obstacle avoidance strategies.After the strategy is determined,according to the designed scene grid map,a mixed path planning algorithm based on grid method and A*method is used to plan the local obstacle avoidance path.Finally,the experiments of different scenarios are designed to test the effectiveness of the intelligent obstacle avoidance algorithm.