Collision Detection Of Human-robot Bounding Box Based On Stereo Camera And Joint Color Information

Modern manufacturing industries are becoming intelligent gradually,traditional industrial robot is fallen.Relatively,the development of applications in the service field has led to the rapid development of collaborative robots and the spring up of human-robot collaboration.The real-time acquisition of human-robot distance and the evaluation of collision contact information in the process of human-robot collaboration are the key points of human-robot collaboration.For the above issues and the case that the color of human arm joint and robot arm joint is single and not similar,a vision-based human-robot collaborative collision detection method was proposed in this thesis,and different types of collision experiments were conducted.The experimental results verified the feasibility of real-time acquisition of human-robot distance and the accuracy of collision information acquired.Firstly,a stereo camera with depth perception was choosed,and it was placed in a fixed position relative to the human-robot.A flexible camera calibration method by shooting chessboard with different angles to calibrate the stereo camera,makes the visually acquired three-dimensional information accurate.Secondly,Camshift was used to track the position of the robot arm joint and the human arm joint based on the color information.To soleve disturbing factors that may cause the target to be lost during the tracking process,such as occlusion and background color similarity,a similarity threshold was set to determine whether tracking process is disturbed or not.Camshift was combined with Kalman filter algorithm for improvement,and the effectiveness of the improved tracking algorithm was proved by the tracking experiment,the robustness of the tracking algorithm was enhanced.Then,a bounding box collision detection method based on space-field was designed.The three-dimensional points corresponding to the tracked joint center points were acquired,their position information were combined with the shape of the robotic arm and human arm,and considering the efficiency of constructing the bounding box,the bounding box of human-robot consist of sphere bounding box and the cylindrical bounding box;The formula of the shortest distance between different bounding boxes was derived,which determines whether to collide,and the specific position of the collision point on the robot arm was calculated.A simulation environment was built by Unity3 D,and the algorithm of collision detection was simulated and verified.Finally,the experiment environment of human-robot collaboration and stereo camera was built,and a test system was developed.The active collision and passive collision tests of human arm robotic arm were conducted.During the experiment,the human-robot distance was obtained effectively in real time,and the distance error calculated by the algorithm was within 0-10 mm when collision occured.Then the collision point at the the link of the robotic arm was calculated,and the result was close to the predetermined point position,and the error was within 10%.Results of experiment showed that the method can effectively evaluate the distance of humanrobot and collision information during the collaboration process.