Robot Collision Avoidance Research Based On Comprehensive Surveillance Video And Kinect

With the development and improvement of technology,the robotics has been a comprehensive subject.The application of the robotics has covered various fields.Although the application fields of robotics are gradually widening,its further development is restricted by the security issue.There may be some collisions of the moving robot with human or obstacles taking place in the workplace,which will threaten the personal safety of workers and cause property loss.The method in this paper belongs to pre-collision control methods,and the research target is detecting the potential collisions in time.After distinguishing the obstacles and extracting the positions of the obstacles,a series of corresponding control measures will be taken to ensure robot safety based on the kinematics characteristics of the industrial robot.The control system in this article is designed based on ABB robot with six degrees of freedom.It combines the global video and the human skeleton detection technology of Kinect.Furthermore,the Kinect is combined with the Kalman filter to predict the postures of the human joints,and provide a good basis for verifying the effective control measures.First of all,this article makes use of the human skeleton detection technology to detect the human location information accurately.The information include the 3D values of limbs,head and bosom are combined with the Kalman filter to predict the limbs movements.On the one hand,predicting the coordinate values of limbs in the next frame image by the Kalman filter could improve the collision identification;on the another hand,it could calculate the velocity vector of limbs movements by the iterative algorithm of Kalman filter,and the human joints danger index is put forward based on the human joints velocity.It could judge the degree of collision risks by comparing the human joints danger index with the threshold value.Secondly,the potential collision is detected by the global video.Firstly,this paper established the forward kinematics of the industrial robot by D-H method.Thus,the coordinates of each robot joints can be calculated based on the robot joints angles.Then,the image processing method could detect the obstacles include human body in the robot workspace.At the same time,in the process of detection,the moving robot will be hided based on the robot joints coordinates which are calculated in real-time,so as to distinguish the robot from the obstacles in the images.Lastly,after image mathematic morphology,the outlines in the binarized images will be detected and the rectangle boundaries of the outlines will be calculated.According to the rectangle information,it can exclude a part of image interference,and then improve the accuracy in judging the obstacles.Finally,based on the information provided above and combined with the kinematic characteristic of the industrial robot,a series of security control measures are put forward according to collisions of robot with different parts of human.The software for image processing and robot motion control is programmed by using VS2010,and the experiments based on the ABB robot have verified that the safety control measures are effective and real-time.