Research & Implementation Of Obstacle Detection Algorithm Based On Feature Points Matching

As an important component of the parking assistance system (PAS), Rear obstacle detection algorithm based on vision is a relatively new study field. At present,there are two major schemes about Obstacle Detection Based on Monocular Vision,Including: Feature-based detection algorithm and Motion-based detection algorithm.However,the first method makes use of the fecture of the obstacle to detect obstacle and need to be informed of the characteristics of obstacle detection,so it only applies to the detection of specific types of obstacles. When the texture of road is rich, the later method will have a lot of false detection and be more sensitive to light changes.In this paper, the traditional detection and tracking methods of feature points are improved. Firstly feature points are detected by Harris corner detection operator,and then using Forstner operator and bilinear interpolation, an iterative methods have been applied to obtain sub-pixel precision coordinates of the feature points that have been detected. Finaly, the corresponding feature points of the adjacent images.are gotten through the optical flow tracking method based on pyramid.This paper presents a new algorithm for obstacle detection:that is based on the displacement of coordinates obtained from projecting the matching feature points on road surface. The displacements of coordinates are calculated by the inverse perspective projection transformation, furthermore, the estimated displacement of camera can be gotten through the voting method. Finally, the threshold is based on the function with the displacement of camera as its independent variables. Judg the coordinate displacement by this threshold function which makes the displacement of camera as independent variable. And then distinguish the features which belong to the obstacle or the road. At last the true height information of these feature points can be obtained.The experiment results show that this algorithm has high detection precision, less calculation, no sensitive to change of illumination, detecting obstacle in any shape, adaptive thresholds and etc. Integrating with other algorithms, this algorithm can enhance the detection rate, also be able to remove the error detection effectively and determine the road surface area.