Vision Based Road Detection And Tracking Algorithm Research

As a main topic of Cognitive Computing Visual and Auditory Information the Major Research Plan of NSFC,Unmanned Drive Technology based on intelligent vehicle platform not only has broad prospects in Drive-Assist applications and Intelligent Transportation Systems,but also stands for the overall strength of the national information processing technology and related industries.As the core content of Unmanned Drive Technology,computer vision based road detection algorithm facing multiple challenges such as accuracy,robustness and real-time property,thus become a focus in the study of experts and scholars.In the research of road detection algorithm,vanishing point estimation based road detection algorithm attracted much attention for its strong applicability when dealing with complex environment road image.However,the existing vanishing point estimation based algorithms remains problems such as of large computing amount,situation that the vanishing point outside the image can't be correctly handled and information among video frames lack of use.In this paper,aimed at such problems,proposed a generalized vanishing point based and the domain orientation constraint road detection algorithm and applied a feature fusion mean shift algorithm for road tracking.The main work is as follows:Aimed at the problem that existing vanishing point estimation based road detection algorithm of large computing amount and situations that the vanishing point outside the image can't be handled,this paper introduces the concept of generalized vanishing point and combine the domain orientation constraint to improve it.The improved algorithm includes three aspects:1)by optimizing the voting strategy,constant vote radius can cover more valid voters,improved the reliability of the vanishing point estimation based algorithm;2)in the domain orientation restraint conditions,the image filter out redundant pixels,reduced the computational algorithms;3)by introducing the concept of generalized vanishing point,determining whether there is a vanishing point in the internal image to select the appropriate road boundary fitting method to enhancement algorithm adaptability.Aimed at the shortcoming that existing road vanishing point based road detection algorithm has less support for video streaming processing or continuous images,this paper applied mean shift algorithm to road tracking,strengthening using of information among frames.Based on the Mean Shift theory,introduced the texture feature into the framework of Mean Shift algorithm combines color features to enhance the algorithm robustness of complex road environment.The algorithm on reference of last frame detection result,narrow the scope of road detection and enhance the real-time capability of tracking algorithm.In order to verify the effectiveness of the improved algorithm,the paper implemented these algorithms and comparative analysis of the proposed algorithm with the traditional road detection algorithm is given.Qualitative and quantitative analysis of the experiment showed that the time efficiency and robustness of generalized vanishing point based and domain orientation constrain road detection algorithm has greatly improved in complex environment,the real-time capability of color and texture features fusion mean shift tracking algorithm greatly improved in the processing video stream or continuous images.Therefore,this paper's work has certain theoretical significance and practical value on Unmanned Drive Technology research.