Research&Implementation Of Obstacle Detection Based On Depth Recovery Technology

As the living standards’improvement, there are more and more vehicles in our country recently. As a result, the traffic accidents have become increasingly prominent problem. Among the various kinds of the traffic accidents, the problems result from the parking assist system, therefore, a considerable number of various auxiliary driving skills become a hot research in the automotive industry. In the wide range of the parking assist systems, people have much more concerned on the systems based on the monocular vision, which have a lot advantages, such as having an intuitive view, large detection area, getting amount of environment information about the back of the car and so on. Naturally, research on the obstacle detection algorithms applying on this kind of assist products is becoming hot.To ensure to help the safety driving, the assist products have to detect any types obstacles in the dangerous areas around the vehicle and remind the driver immediately in the process of parking. According to these application requirements, the paper has mentioned an obstacle detection algorithm based on the depth recovery algorithm. Firstly, preprocess the input image, including reducing noises, improving the contracts and enhance the details. These operations are beneficial to the next steps. Secondly, build the ego vehicle motion model; search the corresponding points in the latest adjacent multi-frames for each pixel in the current frame, using the ego-vehicle motion parameters obtaining from the motion sensor, and model each pixel with their corresponding points.Then, take advantage of the difference between the obstacles’and road planes’motion features to detect the each pixel’s property; if the pixel isn’t consistent with its distribution model, we can define that it is a point in the obstacle, or a point on the road; so the obstacle regions are preliminary detected now. At last, detect the feature points in the tow adjacent images’obstacle regions separately, using the SURF algorithm; and match the feature points using the exhaustion algorithm and obtain a set of precise feature point pairs. And reconstruct for the obstacles according to the depth recovery principle, then remove the areas not meeting the definition of the dangerous ranges. The real obstacle regions are detected at last. This assist product can detect all kinds of obstacles around the vehicle in the complex environment, and then remind the driver the dangers through displaying the detection results on the monitor before him. Additionally, it can also compute the depth of the detected obstacles. After tests in the real environments, the proposed algorithm in this paper has achieved expected effects in recognition rate and performance perspective.