Detecting Untraversable Regions for Navigating Mobile Robot on Pedestrian Lanes

摘要

The abstract should summarize the contents of the paper using at least 70 and at most 150 words. It will be set in 9-point font size and be inset 1.0 cm from the right and left margins. There will be two blank lines before and after the Abstract. ... Pedestrian lane following for autonomous navigation without pre-built maps in unconstrained outdoor environments is investigated. We extend the occupancy grid based mapping method which is dominantly applied in indoor environments to more open outdoor environments with far fewer surrounding objects. A stereo camera with middle ranging capability is used to capture color and depth data of road scenes. The color images are semantically parsed into different regions by using a pre-trained convolutional neural network (CNN). By taking advantages of the depth data, all 3D points of the image regions with non-road class labels are determined as untraversable ones and processed into pseudo laser scans. The pseudo laser scans are further projected into obstacle cells of the local occupancy grid map. By proposing a virtual wall and following it, the robot can autonomously move along the unconstrained pedestrian lanes and keep to the right. We test the navigation system in outdoor and indoor environments using the TurtleBot2 platform. The good performance of occupancy grid mapping and the effectiveness of lane following are validated by the blind test results.