Research On Algorithm Of Object Detection And Path Planning Of Industrial Robot Based On 3D Vision

With the rapid development of science and technology,intelligent manufacturing has become the mainstream trend of the current manufacturing industry.Industrial robots are one of the main driving forces of intelligent manufacturing.However,traditional industrial robots can only do specific jobs in specific positions,and cannot have a variety of different working capabilities like humans.In view of the above defects,this thesis designs a perception industrial robot based on 3D vision,which can identify target objects and obstacles,and autonomously plan obstacle avoidance paths,which can adapt to different working environments and work contents.The specific contents are as follows:First of all,by adding a depth camera to the robot,the robot has the ability to perceive the environment,and then identify the target objects and obstacles through the target detection algorithm.In view of the slow reasoning speed of the YOLOX target detection algorithm,some optimizations and improvements were made to it,and the YOLOX-DBB algorithm was proposed.The algorithm preliminarily improves the running speed of the network by optimizing the Focus module and the SPP module in the network.At the same time,it uses diversified branch blocks to reconstruct the backbone and neck of the network in the training phase,and fuses the diverse branch blocks into traditional The 3×3 convolutional block.This method can reduce the branch structure in the network,which enables the network to speed up the inference speed while improving the detection accuracy.The experimental results on the VOC_2007-2012 data set show that the reasoning speed of YOLOX-DBB is accelerated by17.45%,m AP_0.5 is increased by 1.02%,and m AP_0.5:0.95 is increased by 3.47%.Secondly,the motion path of the robot is obtained through the obstacle avoidance path planning algorithm,and the motion trajectory is generated.Aiming at the slow convergence speed and redundant paths of RRT*algorithm,this thesis designs RRT*-Bezier algorithm.The algorithm speeds up the convergence speed of the RRT*algorithm by introducing the target bias probability;and solves the path redundancy problem of the RRT*algorithm by simplifying the path and eliminating intermediate redundant points;finally,this thesis introduces a third-order Bezier curve Smooth the path to make the robot run more stably.The simulation experiment proves that the improved algorithm not only speeds up the convergence speed,but also makes the final path approximate to a smooth curve.Finally,take the autonomous recognition and grasping task as an example to verify whether the robot can work normally.Based on the SIASUN SR7CL industrial robot,it is equipped with a Realsense depth camera and grippers.The target object is identified by the YOLOX-DBB algorithm,and the object is positioned based on the depth information.According to the relative position of the robot and the target object,use the RRT*-Bezier algorithm to plan the obstacle avoidance path and generate the motion trajectory,control the robot to move to the target position,and complete the grasping task.