Research & Implementation Of Backover Obstacle Detection Algorithm Based On Monocular Vision

With the steady growth of world economy and the continuous decline in car prices in recent years, the holdings of private cars increased significantly. By enjoying the convenience taken form the car, the problems due to traffic accidents also have become increasingly prominent.30 percent of accidents are caused by car's back blind spot when drivers parking. Therefore, it is essential to identify the surrounding environment's useful information accurately. However, most of the existing parking assist products are based on radar or ultrasonic sensor, it has several weak points, such as small detection area, blind spots, having a bad performance in detecting pedestrians, especially children and other small objects. Taking the cost, power consumption, equipment installation, camera demarcating into account, obstacle detection system based on the combination of monocular vision and ultrasonic sensors or radar is direction of development in the future.This paper presents a robust obstacle detection algorithm based on ego-motion estimation. This method establishes a model of ego-motion, gives the membership of image blocks belongs to road(or a solid) by the probability method, then find the motion parameter which have largest probability between the corresponding road image block in two consecutive images by traversing the search space of motion parameters, at last accurately estimates the ego motion parameters.This paper stands on the base of estimating ego-motion parameters, takes the advantage of the difference between obstacle and road plane motion, eliminates the noise points and extracts the obstacle regional in difference image based on motion compensation, uses the model expressed by "Russian box's" obstacle to illustrate the obstacle in image, achieve the robust obstacle detection in any shape, any size, any motion style and speed at last.The results show that the algorithm can support assisting backover in complex scenes, achieve the robust obstacle detection in any shape, any size and no mater it is shop or not, such as vehicles, pedestrians, ball, poles and so on. At the same time, the express model of any obstacle established by this algorithm illustrates the shape location of the obstacle directly, provides an important basis for avoid collision algorithm in assist park system.