Research And Implementation On Multi-camera Based2D Round-view Image Building

Increasingly improved with people’s living standards and the road construction,an increasing number of private cars enter in family of ordinary people. Along withfrequently accidents occurred because of the blind zones of vehicles, Car enthusiastsand the field of car driving assistance have paid more and more attention to theproblem of vehicle bland area. This paper is based on the requirement of an ongoingresearch of Intelligent Car Driving Assistant project, and aimed to study andimplement one kind of2D round-view system with multiple cameras, so asto completely eliminate the blind zone around the vehicle body and avoid theoccurrence of safety accidents in the vehicle driving, backing and turning.The multiple-camera based round view generating system make full use of thefour cameras installed around a car to realize real-time monitoring for theenvironment around the vehicle, furthermore, by comprehensively processing theimages of different cameras, it will generate a360°round-view composite imagewith vehicle body centered. After the composite image sequence been transfer toauto cab, it will be convenient for the driver to know real-time environmentand traffic condition and to make the right decision.In this paper, after in-depth study of the development situation and relatedtechnology in this field and, the principles of the multi-channel image acquisition tothe two-dimensional ring view generation are summarized, and the problems to dealwith in every step of the process realization are discussed. In order to generate anideal2D360°round-view image, processing methods like image acquisition, imagepreprocessing, camera calibration and joint calibration, round-view stitchingand image fusion, etc. are keys to system implementation.On basis of the four cameras, we did some research and Implementation of the 2D round-view image generating system, and main research contents are as follows:Through the study of imaging model of fisheye lens and analysis of common methodsto proofread fisheye images, an improved bilinear interpolation based sphericalcoordinate positioning method is raised, which comes a better correction of the imageedge distortion. In order to produce a round-view image with multiple-image fusion, itis proposed to use the local template region method to segment the final round-viewimage in advance, which can effectively speed up the stitching process of multipleimages. In order to generate overlooking perspective round-view image, afterstudying different transformation method, a method of the Harris corner extractionbased overlooking transformation matrix calculation is raised, which effectivelyreduces the error caused by Manual points picking. Finally, considering the real-timeprocessing performance of the multiple videos, control problems, OpenMP multi-coreprogramming technology based parallel optimization method is put forward in oursystem, so that the processing speed is well improved.The experimental results show that, in a certain time, our method can generate anideal2D360°round-view image with vehicle body centered.