Research On Vision Based Mobile Robot Obstacle Avoidance Algorithm

Recently,with the development of the robot technology,mobile robots have been widely used in different fields.Among them,the mobile robot obstacle avoidance technology,as an integral part of the mobile robot field,has received more and more attention,and has become a research hotspot in current industrial and academic field.However,the two key technologies in mobile robot obstacle avoidance,obstacle object detection and path planning,still face many problems to be solved.For the obstacle object detection,there exists the problems such as low detection efficiency,low detection accuracy,and overly high computational cost of detection algorithm,resulting in that the mobile robots are unable to perform real-time detection in the actual environment.For path planning,there exists the problems such as low searching path efficiency,long path distance,large number of turns and angles,and unsmooth path.In this paper,we focus on the object detection and path planning technical issues involved in mobile robot obstacle avoidance technology,and the main research work and results are as follows.To solve the problem of limited computing resources in the obstacle detection technology based on deep learning,a model clipping method that combines the reconstruction parameter γ in the BN(Batch Normalization)layer and the convolution kernel weight has been proposed to perform the original YOLO-V3 target detection network pruning.Firstly,the original YOLO-V3 target detection network has been sparsely pruned based on the BN layer,and the model and weight parameters retained after the sparse pruning have been pruned based on the weight of the convolution kernel.Secondly,the pruning network had been fine-tuned and re-trained.After training,an obstacle detection network has been obtained for obstacle avoidance of mobile robots.Finally,the image binarization processing method has been used to distinguish the detected obstacles in the foreground from those in the background.The experimental test on the COCO target detection data set has proved that the deep learning target detection pruning network proposed in this paper can simplify the original network while ensuring the detection accuracy,and effectively solve the mobile robot computing resources restricted issues.To solve the problems of low efficiency of searching path,long path distance,large number of turns and angles,and unsmooth path of mobile robot obstacle avoidance and path planning,an optimization method of variable proportion heuristic function based on A*path planning algorithm has been proposed together with a smooth optimization method by traversing the original path.Firstly,the heuristic function in the A*path planning algorithm has been optimized to reduce the number of path search nodes and improve the path search efficiency.Secondly,the search path has been optimized in three ways to smooth the obstacle avoidance path of the mobile robot.Finally,the simulation experiment has proved that the path optimization method proposed in this paper can solve the problems of low efficiency and smooth travel of the mobile robot in the obstacle avoidance process,further proving the advancement and effectiveness of the method proposed in this paper.