Research And Implementation Of Moving Obstacle Detection Technology Based On Monocular Vision

With the rapid economic growth and continuous promotion of the car, the number of the private car increases dramatically, as a result, the traffic accident which is caused by parking also occurs frequently. Therefore, advanced parking assistant system with anti-collision warning function attracts people's extensively attention. As one of the most important techniques in parking assistant system, the detection of obstacles (particularly moving obstacles), which are at the back part of the vehicle, plays an active role in improving vehicle's security performance and reducing traffic accidents, and may have a broad prospect.The research topic of the thesis is on-board monocular vision-based moving obstacles (which are at the back part of the vehicle) detection technology. It consists of three parts, in the first part, the existing corner detection algorithms are classified, two classic algorithms (Harris algorithm and SUSAN algorithm) are implemented and contrasted, which laid a solid foundation for the calculation of optical flow and image matching. The second part mainly focuses on the optical flow algorithms which are based on gradient and matching. A consistent data check technology is proposed, which makes the result of the optical flow more accuracy. The third part of the discussion focuses on moving obstacles detection under the circumstances of stationary vehicle and moving vehicle.In case of stationary vehicle, a based on three-image background recovery method is proposed; In case of moving vehicle, an optical flow-based algorithm is proposed and implemented, and the detection of non-rigid body movement is effective.Experiments show that the algorithm mentioned above could get a satisfactory result and is useful for moving obstacles detection and their movement characteristic classification.