Texture-based Road Vanishing Point Detection

Road vanishing point detection plays an important role in computer vision andIntelligent Transportation System.It analyzes the videos or imagesfrom the sensorsettled in front of the intelligent vehicle, and detects the position of the vanishing pointformed by parallel lanes and other straight lines parallel to the lanes. Therefore, how todetect road vanishing points in various complex road scenes rapidly and accurately isthe key to solve the problem. This paper mainly studies vanishing point detectionalgorithms based on statistical estimation and introduces Weber Local Descriptor topropose an effective road vanishing point detection algorithm.Firstly, this paper expounds the background and research significance of roadvanishing point detection, summarizes existing research status and the main difficultproblems of vanishing point detection.Secondly, this paper introduces the theory of perspective projection and analysis ofthe mathematical model of vanishing points. It introduces space-conversion-technologybased, points-of-intersectionbased and statistical-estimation based vanishing pointdetection methods in detail and analyzes the advantages and disadvantages of thesemethods. Among them, vanishing point detection based on statistical estimation areintroduced emphatically, especially vanishing point detection algorithm based on localtexture descriptors and voting scheme.Thirdly, Weber Local Descriptor is introduced in detailand is improved in theapplication of road vanishing point detection. For the differential excitation component,this paper defines a new calculation formula from the weber’s lawand introduces squareroot to make differential excitation component have stronger ability to distinguish smallvalues; For the orientation component, this paper describes the texture direction throughGabor filter banks, which can effectively utilizes information of all pixels in theneighborhood; For multi-scale expansion, this paper effectively utilizesinformation ofall pixels in the neighborhood, making the descriptor more accurate.Finally, this paper proposes a vanishing point detection method based on theimproved WLD. First of all, image preprocessing reduces the effect of noise invanishing point detection process. Then, the improved WLD is introduced to extractlocal texture feature. Then, differential excitation componentin the improved WLD isused to extract road salient regions. Finally, all pixels in salient regions vote for road vanishing point according to orientationcomponent in the improved WLD and theproposed linear voting scheme.A large number of experimental tests show that theproposedmethodcandetect vanishing point accurately and rapidly, which can basicallymeet the needs of real-time processing.