Human-robot Collision Detection Based on the Improved Camshift Algorithm and Bounding Box

Aiming at the problem of collision detection and collision point information evaluation in the process of human-robot collaboration, the binocular camera is used as an external sensor to observe. Collision detection is realized by tracking the motion of human-robot through the color information of joints. Firstly, the Camshift algorithm is used to track the position of the manipulator joints and the human arm joints based on the color information. In order to solve the factors that may cause target loss during tracking process, such as shelter and background color similar problems, Kalman filter is integrated on the basis of Camshift algorithm. A similarity threshold is set to judge whether there is interference in the tracking process. The tracking experiment proved that the Kalman filter is effective and enhances the robustness of the tracking algorithm. Secondly, a bounding box collision detection method based on space domain is designed. The sphere bounding box and the cylindrical bounding box is used as the human-robot bounding boxes. The equations of the distance between different boxes are derived and the position of the collision point on the manipulator is calculated. Finally, an experimental environment is built for verification. The distance error of the collision is within 0–10 mm, and the position error between the calculated collision point and the pre-determined collision point is within 10%.