Stereo vision-based obstacle avoidance module on 3D point cloud data

This paper deals in building a 3D vision-based obstacle avoidance and navigation. In order for an autonomous system to work in real life condition, a capability of gaining surrounding environment data, interpret the data and take appropriate action is needed. One of the required capability in this matter for an autonomous system is a capability to navigate cluttered, unorganized environment and avoiding collision with any present obstacle, defined as any data with vertical orientation and able to take decision when environment update exist. Proposed in this work are two-step strategy of extracting the obstacle position and orientation from point cloud data using plane based segmentation and the resultant segmentation are mapped based on obstacle point position relative to camera using occupancy grid map to acquire obstacle cluster position and recorded the occupancy grid map for future use and global navigation, obstacle position gained in grid map is used to plan the navigation path towards target goal without going through obstacle position and modify the navigation path to avoid collision when environment update is present or platform movement is not aligned with navigation path based on timed elastic band method