Development and Characterization of a Safety System for Robotic Cells Based on Multiple Time of Flight (TOF) Cameras and Point Cloud Analysis

In this paper, a vision system for safety applications in human-robot collaboration is presented. The system is based on two Time-Of-Flight (TOF) cameras for 3D acquisition. The point clouds are registered in a common reference system, and human and robot recognition are then implemented. Human recognition is performed using a customized version of the Histogram of Oriented Gradient (HOG) algorithm. Robot recognition is achieved using a procedure based on the Kanade-Lucas-Tomasi (KLT) algorithm. Two safety strategies have been developed. The first one is based on the definition of suitable comfort zones of both the operator and the robot; the second implements virtual barriers between the operator and the robot. The vision system has been characterized in terms of (i) human and robot recognition performance, (ii) correctness of the detection of safety situations and (iii) evaluation of the time delays in the detection. The results show that the human operator is robustly recognized provided that he moves frontally with respect to the TOF cameras and the robot is always recognized. The safety situations are always identified correctly with an average time delay of 0.86 0.63 s (k=1).