Estimation of Human Position and Velocity in Collaborative Robot System Using Visual Object Detection Algorithm and Kalman Filter

Safety issues are increasing as collaborative robots and people share workspaces. In this paper, we propose a technique for estimating a single human position and velocity using two fixed RGB cameras. To detect human, an object detection algorithm composed of convolution neural network is used. The detection area in images obtained from the detection algorithm are used to calculate the human position in the experimental environment with some partial visual obstruction through coordinate transformation. Then we use Kalman filter to estimate the filtered position and velocity. Finally, we suggest how to predict the position and velocity when human is blocked from the cameras due to obstacles.