Vision Based Intelligent Vehicle Autonomous Navigation

Nowadays, intelligent vehicles which are devoted to enhance traffic efficiency and safety in urban environment are given more and more attention. The autonomous navigation technology is one of the core technology for intelligent vehicles; Vision based autonomous navigation technology is a promising research direction in the field of intelligent vehicle. Therefore, vision based autonomous navigation method for intelligent vehicle has been researched.Research work is carried out center on three basic parts of vision based autonomous navigation method, i.e. camera calibration method, lane detection and vehicle localization method, vehicle control method, and an extended part, i.e. pedestrian detection.An important problem with respect to camera calibration part is how to calibrate camera distortion; another important problem is how to realize vehicle coordinate-camera calibration. For the first problem, the derivation of camera model is introduced detailedly. A new camera distortion calibration method based on optimization with dimension reduced by IPM (Inverse Perspective Mapping) is proposed, Experiments on both synthetic data and real data validate the efficiency and accuracy of the proposed new camera distortion calibration method. For the second problem, a convenient calibration method based on grid pattern is proposed, and tested by real experiments.The key problem with respect to lane detection is how to detect the two edges of a lane on image. For this problem, a lane detection method is proposed. First, several candidate line segments are sifted out using Hough transform; then the edges of the lane are extracted using parallel constraint and topological constraint. Real experiment shows the efficiency of the proposed lane detection method. For vehicle localization, vision localization method as well as how to fuse DR (dead reckoning) data with vision localization result using EKF is introduced detailedly.The core part of vehicle control is vehicle lateral control. A basic problem is how to realize stable and converged lateral control. There are many existing methods to tackle this problem; a survey on existing lateral control methods is given. A further problem is how to improve comprehensive control effect. For this problem, A new lateral control method named adaptive predictive control is proposed. Experiments show that the proposed method not only is stable and convergeable but also has considerable enhancement on comprehensive control effect, compared to the state proportional feedback method which is a commonly used traditional lateral control method.The problem of pedestrian detection is how to realize feasible and reliable detection on pedestrians in road environment. Two classes of traditional methods are vision based method and range data based method. Although each of these two classes of methods has its own advantages, they also have their apparent disadvantages, thus it is difficult for each kind of method to guarantee both feasibility and reliability. For this problem, a new method of camera and laser radar co-detection is presented. The innovative points mainly lie in the cooperation of camera and laser radar which demonstrates both the advantage of laser radar on"detection"and the advantage of camera on"classification". Experiments validate the efficiency and reliability of the proposed co-detection method. As a part of the co-detection method, the proposed camera and laser radar co-calibration method is also an innovative point.Vision based intelligent vehicle autonomous navigation is successfully realized by integrating the research results of above four parts.