Vision based collision avoidance by plotting a virtual obstacle on depth map

Obstacle avoidance is an important task for autonomous navigation. The paper presents a simple vision based obstacle detection and avoidance algorithm for mobile robots. Using a monocular vision system, a depth map is created in which every pixel of the image captured by the robot is classified in one of the two categories- ground or obstacle. A virtual triangular obstacle is then created on the depth map. To avoid this obstacle, the robot is directed along its edge having the maximum magnitude of slope. A distinct feature of the proposed algorithm is that along with the position, it takes into account, how densely populated are the obstacles in the field of view and turns the robot in the direction offering less hindrance. The algorithm can be effectively deployed for situations involving environment exploration. The algorithm has been validated on a differential steering robot with an on-board camera, driven by a remote computer which guides the robot, based on the algorithm implemented in MATLAB®. The algorithm is tested under uniform indoor lighting conditions and avoids collision of obstacles in its path successfully.