Research On Scene Reproduction Technology Oriented To Visual Parking Assist System

With the development of camera manufacturing technology,the improvement of embedded device computing power and the continuous improvement of people’s demand for active safety of automobiles,vision-based assisted parking systems have been more and more widely used.At present,the reversing imaging system has been widely popularized,but the two-dimensional panoramic system and the three-dimensional panoramic surround view system are currently only assembled on higher-end models.Research on the visual parking assist system can accelerate the popularization of visual parking assist,eliminate the blind spots of the driver’s vision,and improve the safety of parking.This paper first analyzes each camera model,studies the calibration method of the camera’s internal and external parameters,improves the checkerboard corner extraction algorithm,and uses the initial internal parameters to accelerate the iteration of the calibration algorithm to achieve rapid calibration of the vehicle camera.Then,by calculating the mapping relationship between the pixels of the original picture taken by the camera and the pixels of the top view,a projection table is generated,which realizes the rapid generation of a top view based on the original image.Finally,aiming at the stitching seam and chromatic aberration in the top view,the algorithm of stitching elimination and image color adjustment is studied,and the optimization of the top view is realized.Aiming at the panoramic view algorithm,this article first introduces the modeling method of 3D surface in the 3Ds Max environment,and saves the 3D surface to the local in the format of obj.Then,the vertex texture coordinates are calculated through the projection relationship from the surface vertices to the image pixel coordinates,and the texture coordinates are saved.Finally,Open GL is used to achieve simultaneous rendering of three-dimensional surfaces and vehicles,and methods such as viewpoint transformation and back culling are used to complete scene images under different viewpoints and complete the panoramic surround view algorithm.Aiming at the defect that the parking assist system cannot perceive the bottom of the vehicle,this paper proposes a transparent vehicle bottom algorithm,which is based on the image taken by the on-board camera and the state of the vehicle to realize the estimation of the bottom image.This article first studies the method of reading and parsing vehicle CAN messages using C++ on Linux platform.Then the vehicle is kinematically modeled,and the vehicle underside image extraction algorithm is completed using multi-threading and other technologies,and the transparent underside algorithm is optimized for defects such as excessively large stitching of the vehicle underside image.Finally,experiments are conducted on all functions of the visual parking assist system.The experimental results show that the visual parking assist system proposed in this paper is reasonable in design and effective in use,and can well eliminate the stitching of the overhead image and the chromatic aberration of the images taken by different cameras.,And through the transparent undercarriage algorithm,the inference of undercarriage images can be realized,and parking safety can be improved.