A moving object and obstacle detection system in real-time AVG and AMR applications

One of well-known problems faced in Automatic Vehicles Guidance (AVG) and Autonomous Mobile Robots (AMR) applications is the moving object and obstacle detection (MOOD). The paper deals with this topic by a real-time system in a stereo vision framework. The system employs the best methods in literature and opportunely modifies them to obtain the best compromise between the high frame-rate and the high accuracy requirements. A disparity map is calculated to have a 3D representation of the scene and to recognize obstacles. An efficient algorithm for motion vector analysis, based on optical flow, is used to segment moving objects and obstacles. Results are presented with reference to a synthetic database created ad hoc to evidence some interesting cases of object / obstacle trajectories.