Research On Parking Assistant System Based On Panoramic Stitching

As a new technology to help the driver safely parking,the panoramic parking assistant system has become a hot spot in the field of automotive safety technology research.The system can provide a real-time bird's-eye panoramic view around the vehicle body for the driver,which effectively reduce the incidence of traffic accidents caused by vehicle blind spot.This paper focuses on designing the algorithm of the panoramic parking assistance system.Firstly,by analyzing the camera calibration algorithm and the fish-eye distortion model,Fish-eye camera is calibrated with Zhengyou Zhang calibration method,which can get high-precision intrinsic parameters and distortion coefficients.Secondly,In order to solve the problem of some effective pixels loss in fisheye correction images,an improved undistorted algorithm based on fisheye calibration is proposed.The experimental results show that the proposed algorithm can correct the distorted image better than other algorithms.We adopt vignetting bi-directional filtering algorithm to eliminate the vignetting effect in the corrected images and then use direct line transformation to obtain top view images.After that,to avoid the secondary distortion of the stitched image which is caused by distortion in the marked point area,an improved image stitching algorithm based on marker points is raised.The test results verify the stitching effect of the proposed algorithm is better than the general algorithms.Finally,a fade-in and fade-out image fusion algorithm is used to gain seamless mosaic image.At the same time,according to relationship of the coordinates between the pixels in the original fish-eye image and corresponding pixels in the panoramic image,a panoramic parameter map generation algorithm is designed to quickly generate panoramic bird's-eye view image and improve the real-time performance of the system.The algorithm of the panoramic parking assistance system is validated by collecting data from constructing test environment on the car.The results show that the designed algorithm has good real-time performance and effectiveness,which satisfy the requirements of system,and support a variety of resolution video and can be suitable for different car models.