Image Stitching Algorithm Research Based On Driverless Car

Environment perception is a key technology for autonomous driving of driverless car toachieve. Image information has the advantage of low-cost to realize the vehicle environmentperception. Research on multi-source multi-view images mosaic is helpful to realize the fullrange of perception of driverless car environment. Driverless car in the detection andidentification of the outside world information is interested in different regions, in order tocomprehensive traffic scene perception, this dissertation studies multi-source multi-viewimage matching algorithms.Based on the purpose of acquisition ground lane for driverless car, this dissertationpresents a method for images mosaic algorithm of wide-angle camera and telephoto camera.Image mosaicrealizationofmulti focal lengthcansolvedriverlesscar indetectingthegroundinformationwhenthe contradictionbetweenrange and distance. Moreover，the algorithmofmulti-view image stitching, solves the problem in detail and overall to coexistence in largescale complex traffic scenes. The mainly study results of this dissertation are as follows:Firstly, an improved configuration mode of visual sensors is presented in thisdissertation, which based on the current status of driverless car sensor configuration, andcomprehensive the requirements of driverless car when detecting the lane line, traffic lights,traffic signs and pedestrians.Secondly, this dissertation studies images mosaic of multi-focal length and uses it todetect ground signs. The SIFT algorithms is adopted to stitch multi-focal images from wide-angle lens to telephoto lens. And RANSAC method wipes off mismatch, which improves thematching precision. Experimental schemes are designed for different ground information,and analyses the matching results.Finally, as the high requirement of instantaneity when detecting traffic lights, trafficsigns and pedestrians, this dissertationadoptsSURFto match images. For specialapplicationof the driverless car traffic scene, studya wide range of image stitching algorithm for imagescaptured by3telephoto cameras and two wide-angle cameras.