The Research Of Multi-environment Road Detection Algorithm For Autonomous Navigation System

Vision-based Intelligent Navigation System is one of the hot research topics in the field of Computer Vision. In the system, Road Recognition algorithm is hired to provide accurate road information and is an important part of the system. Real-time and robustness are critical problems current road recognition algorithms have to take care of.With the development of Computer Vision, Pattern Recognition and Digital Video technologies, amounts of researchers have made a deep research on the real-time and rubustness of Road Recognition algorithms, and have proposed several new ideas and methods. However, vision-based road recognition is still a challenge since the algorithm has to deal with shadows, bad weather, changing illuminations and so on.Image Segmentation and Vanishing Point detection are two normal technologies usually used in Road Recognition. Among the Image Segementation based Road Recognition algorithms, lots of algorithms segement the road area and non-road area based on single channal of image. This kind of algorithm is usually real-time, but lack of robustness. Inversely, those based on Vanishing Point detection is robust but not real-time. According to current research, How to achieve a good balance between real-time and robustness is the destiny of this paper. After the careful research of main Road Recognition algorithms, this paper proposes two improvements from the points of Image Segmentation and Vanishing Point detection respectively. The main work of this paper is described as follow:Firstly, to solve the problem that most of current algorithm is not robust enough for they make use of single channal information, the paper will make full use of color information, and propose a4D model based color image segmentation algorithm, as well as a Road Recognition algorithm based on it. The new segmentation algorithm introduces4D model to describe statcistic distribution of every pixel to make sure every channal of each pixel is considered as a integrety, which helps to provide accurate and plenty of color information for Road Recognition to set apart road area and Non-road area. Meanwhile, the algorithm also use vector to expand normal Otsu algorithm to4D model, and get a4D Otsu color image segmentation.Secondly, in order to make the Vanishing Point detection based algorithm more real-time, the paper proposes a real-time and robust algorithm from the point of improve the real-time of vanishing point detection. The algorithm firstly preprocess image and abstract useful pixels, then only calculate useful pixels’orientations. This process help to deduce the computation and improve the real-time of the algorithm. During the process of voting for Vanishing Point, my algorithm also has improved the voting rules by introducing the measurement of reliability according to the distance between to pixels. At the same time, my algorithm introduces a new boundary calculation algorithm which combines color, graduate, edge information together. Theses improvements also make my algorithm general to structure and unstructure roads.At the end, this paper adapt the images which are captured by Hunan University Unmanned Ground Vehicle to testify my algorithms, and compare the result of my algorithms with some famous Road Recognition algorithms. The experiment result shows that my algorithms are robust and real-time...