On real-time obstacle avoidance using 3-D point clouds

Obstacle avoidance is a key problem required in the mobile robot navigation technology. In this paper, we build up a 3D environment in an intelligent surveillance robot through registration of point clouds and present a method to deal with the obstacle avoidance problem. Firstly, the framework of the intelligent robot is given. Then we register the point clouds from an RGB-D Kinect by applying an ICP (Iterative Closest Point) algorithm. The online processing is done using the ROS system. Next, we propose an approach to avoid obstacles by projecting 3D data to a plane thus 2D technologies which need less computation can be used. Finally, some experimental results are presented to be valid.