Research On Intelligent Vehicle Autonoumous Navigation In Campus Areas

With the acceleration of urbanization process, urban traffic problems have become more and more serious. Intelligent vehicle technology is drawing attention, for it is an effective solution to solve the urban traffic problems, such as the safety and efficiency problems. Urban environment is the trend of research on intelligent vehicle. It is a semi-structured and complex environment, including the intersection environment, which can not be described by a unified mathematical model. Widely use and development in urban environment of traditional navigation methods, which based on high-accuracy GPS, vision or 3D laser, are limited, due to their own inherent limitations are limited widely use and development. These limitations include: One navigation methods can not cover all the navigation area; the cost of sensors can hardly be afforded; the real-time and reliability request can not be well satisfied. Therefore, to find a solution that can solve the three main issues is so significant to the application and development of intelligent vehicle.Campus environment is the object of the research, while to solve these three existing problems is the goal of the research. Campus environment is a typical semi-structured urban traffic environment. It is one of the stage research objects before intelligent vehicle can be applied in urban environment. The research consists chiefly of system design, navigation in the road area and navigation in the intersection area. As follows:(1) The section of system design analyzes the features of autonomous navigation environment, and puts forward the navigation method based on the idea of environment division; It presents the configuration of intelligent vehicle system. For the first part, the navigation environment is divided into road environment and intersection environment. Different methods are used to the two environments based on their own features. For the road environment, road border tracking method that based on vision is used; for the intersection environment, since its structure is irregular, a method based on multi-sensor fusion is used. Division of navigation environment makes the autonomous navigation no longer required one same method to cover all the navigation area. For the second part, namely, the configuration of system is based on the navigation needs. The intelligent vehicle uses camera on road detection and signs detection; it uses laser for collision avoidance; it uses low-accuracy GPS's information for environment division; it uses encoder to solve the intersection problem.(2) The section of navigation in the road area, proposes a reliable and real-time road detection algorithm; it shows how the preview-following control method works on vehicle lateral control; it presents a filtering method for rotation angles of vehicle control output. For the first part, the road detection process combines the color analysis and edge detection. Environment noise filtering is conducted by analysis of RGB color space, while the road boundary's contours are got through the edge detection. Then, the road boundary is obtained from Hough transform and a set of constraints. Furthermore, a method about on generating the region of interest is given, and it can help to reduce the image processing area. Experiments on various real scenes show the efficiency of the proposed method. For the second part, the ways how to apply preview-following control method on the two base cases (both or only one boundary has been detected) is presented. For the third part, it proposed an improved filtering algorithm which aims to improve the traditional clipping-averaging filter's over-reliance on the selection of filter threshold. Experiments shows the improved filter will not fail even when the filter threshold is poorly selected, while maintaining the filtering performance.(3) The section of navigation in the intersection area, presents a way on how to divide the intersection navigation problem into subdivision and how to solve the sub-problem; it gives a judgment condition about collision avoidance which based on vehicle model. For the first part, the intersection problem is divided into five sub-problems, which are intersection detection, local initially positing, target position and orientation computing, path planning, and real-time positioning. The research describes solution on these five problems in detail. It fuses multiple sensors'information, which complement each other, to achieve a low-cost solution on intersection problem. Experimental results, in the real situation, prove that the method is efficient. For the second part, it gives a judgment condition about collision avoidance. It is based on vehicle model, while the traditional method only consider whether if there are obstacles within the safe distance in front of vehicle. The traditional method is unreasonable when the vehicle is making a turn.