Real-Time Computation of Distance to Dynamic Obstacles With Multiple Depth Sensors

Abstract

We present an efficient method to evaluate distances between dynamic obstacles and a number of points of interests (e.g., placed on the links of a robot) when using multiple depth cameras. A depth-space oriented discretization of the Cartesian space is introduced that represents at best the workspace monitored by a depth camera, including occluded points. A depth grid map can be initialized off line from the arrangement of the multiple depth cameras, and its peculiar search characteristics allows fusing on line the information given by the multiple sensors in a very simple and fast way. The real-time performance of the proposed approach is shown by means of collision avoidance experiments where two Kinect sensors monitor a human-robot coexistence task.