Obstacle Detection Method Based On Single Camera

From 1960s, the first mobile robot-SHAKEY was born, to present, Boston Dynamics invented a new generation BigDog able to walk on icy terrain and recover its balance when kicked from the side. Mobile robot has become a hot topic with its many potential applications.One of main problem for the mobile robot is the obstacle avoidance. With the obstacle avoidance, the mobile robot can move safely at the uncertain environment. This paper introduces a novel obstacle detecting method based on one single camera which is used for the obstacle avoidance.Firstly, a depth learning method is used. It will use a Gaussian MRF Model to present the relationship between the image features and depth. Once the model is known, the predicted depth map (as the output) will be obtained from the new image (as the input) through this model.Secondly, combining with the segmentation image and the predicted depth map, the position of the obstacle at the image is found. Then we can know the deflect angle through the transformation from the image coordinates to the world coordinates. It would help the mobile robot indentify the dangerous direction. When the mobile robot knows the obstacle depth and direction, it can avoid obstacle effectively.Laser scanner is used to collect the depth data for training here. Mechanical structure and hardware is designed to make the data collection more effectively and quickly.Finally, the results show a good performance of this new obstacle detecting algorithm.