Research On Visual Safety Monitoring Method In Human-robot Coexistence Environment

With the widespread introduction of the concept of human-machine collaboration,the cooperation between humans and robots in the production process has become closer.Due to the characteristics of the robot’s fast speed and large torque,once an operation error occurs or the robot loses control,it is likely to cause personal injury and equipment damage.In addition,due to the lack of perception and cognitive ability of robots,it is difficult to detect the existence of people in the surrounding environment,and it is even more difficult to recognize the posture and movement of human bodies.This also makes robots prone to safety hazards when working with people.To address these problems,this paper focuses on the subject of industrial safety in the environment of human-machine coexistence,takes human-machine graphic information recognition,human-machine target online tracking,and human-machine safety rate estimation as the research content,and conducts an online human-machine safety monitoring method based on computer vision.The main contents of this paper are as follows:First,aiming at the problem of online recognition of human-machine targets,a method based on deep learning algorithms to identify human and robot targets in industrial environments is utilized.The human-machine target image data set is established by manual annotation and data enhancement,and the target in the image is recognized by the YOLO R(You Only Learn One Representation)target detection algorithm in the deep learning workstation environment.Experimental results show that the adopted method can accurately and quickly detect objects in a human-machine coexistence environment,and can satisfy online human-machine object detection applications.Secondly,an improved Deep SORT Multi-Object Tracking(MOT)algorithm is proposed for the online tracking of the human body based on video sequences.After the target detection algorithm outputs the bounding box,the appearance features in the bounding box are extracted by a neural network,and then the state vector of the existing trajectory is predicted by the strong tracking module,and the feature vector of the bounding box is matched to the existing target through the Hungarian algorithm.Finally,the trajectory is updated by the strong tracking module.According to the quantitative analysis of the target tracking index and the results of the visualization experiment,the proposed algorithm improves the accuracy and stability of the multi-target tracking algorithm.Then,aiming at the problem of safety estimation in the process of human-machine cooperation,a location-by-detection method is realized by combining RGB-D images,and the safety factors in the environment of human-machine coexistence are extracted.The "safety rate" in the existing safety estimation models based on distance and collision prediction is quantified and modeled respectively,and an improved human-machine safety rate estimation model is proposed by combining the characteristics of each model.The experimental results show that the proposed method can combine the advantages of the distance safety estimation model and the collision prediction safety estimation model to improve the accuracy and robustness of the system.Finally,a safety monitoring system under the environment of human-machine coexistence is constructed,the overall framework of the system,the system software,and the hardware development platform are designed,and the human-machine safety monitoring experiment is carried out.The human-machine safety monitoring system platform is built on the basis of the depth camera and the Robot Operating System(ROS).The constructed system realizes the functions of online object detection,multi-target tracking,safety rate evaluation of humans and robots,and so on.The experimental results show that the safety monitoring system can achieve the expected effect and can be applied to the online monitoring of human-machine safety in the industrial field.