Research On Algorithm In Panoramic Parking Assistance System

As people's living standards continue to improve, more and more cars come into people's life, and car safety driving has become a popular research. Panoramic parking assistance system is proposed to assist the driver in safer parking. Usually we need to install fisheye cameras in the vehicle body in order to implement this system. With these cameras we can capture real-time scene information around the car, and then the panoramic top-view image is displayed on the cab screen in real time after a series of image processing.This paper studies the algorithm implementation of the panoramic parking assistance system, and implements the common algorithm based on four fisheye cameras. Firstly, the fisheye cameras are calibrated by the captured images. Then, the distorted images are rectified by the calibrated camera parameters. Finally, four undistorted images obtained before are transformed into four top-view images which then generate the panorama top-view image through image stitching or mapping. With this procedure we make corresponding experiments and achieve good results.This paper presents a novel method of panoramic parking assistance, which achieves the work of panoramic parking assistance with single camera. The procedure of this method is: firstly, the feature points are detected in previous frame. Then, the detected points are tracked by optical flow tracking algorithm in order to get corresponding points in current frame. Finally, the transformation matrix between previous frame and current frame is obtained according to corresponding point pairs, and with the transformation matrix we produce the panoramic image. In the method, we choose AGAST algorithm to detect features, and we use the pyramid LK optical flow algorithm which is used most commonly to track the features. In order to fit our specific application, we improve the pyramid LK optical flow algorithm using the planar homography constraint. The tracking speed becomes faster.