Research On The Key Technology Of Intelligent Vehicle Navigation System Based On Satellite And Vision

The intelligent vehicle, as an integrative system with the function of environment-sensing, decision-making, self-control and other functions, has been an important part of the Intelligent Transportation System. Embodying the latest research results involving information science and artificial intelligence, its research has become the tactic research object of all countries. Vision navigation is its essential core technology. At the present, the research relates to computer vision, data fusion and so on, is the challenging research topic.As broad camera monitoring area, abundant information, slight multi-system interference, etc, coupled with its continuous improvement of the performance-to-price ratio, the machine vision research has been widespread concerned in recent years. However, due to the constraints of visual information redundancy, easily influenced by environment and processing speed, at present tests, the visual sensors, equipped in the intelligent vehicle, are often used for traffic signal recognition, traffic signs identification and other auxiliary functions. The vehicle environment perception and obstacle detection, are mainly completed by active millimeter-wave radar and laser radar. This kind of sensors is more expensive when long-distance detection obtaining high precision is needed. So it is difficult for widely utilizing. Therefore, for vision-based navigation vehicles, visual nevigation is still the only way to realize the intelligent vehicle.Combined GPS/INU (Inertial Navigation Unit), the global vehicle navigation obtains the vehicle state (position, attitude and speed, etc.) in real-time, but the installation of encoders and inertial sensors is complex. At present, with GPS technology continues to improve as well as GLONASS/BD/GALILEO navigation systems gradually put into use, the precision of GNSS location is significantly improved. Utilizing network RTK technology, the dynamic positioning accuracy can achieve a centimeter-level. Therefore describing the vehicle state (position, attitude and speed, etc.) through the dual GPS composition baseline becomes feasible.In the paper, the key technologies of vision navigation with GNSS technologies are studied. Including dual GPS and stereo vision navigation system components, the external parameter self-calibration, road detection, stereo matching for obstacle detection and obstacle avoidance path planning algorithm.Its concrete content and the results are:1) the composition of the navigation system, automatic acquisition of vehicle attitude, stereo camera model, camera calibration and conversion between coordinate systems were studied. With the actual application, the stereo vision navigation system with dual GPS integrated stereo video camera was constructed. This method is the foundation of vision navigation. Then, automatic gain vehicle attitude in real-time has become feasible. This method not only provides vehicle status information for the vehicles intelligent navigation (position, attitude and speed, etc.), but also can measure the absolute position of scene, through stereo camera. Its structure is simple and data processing is fast. The calibration experiment exhibits that the method can obtain high accuracy, can satisfy the requirements of intelligent vehicle navigation.2) Working in accurate and timely detection of road area in a complex environment, a real-time and robust road detection method is presented, which is based on priori knowledge database and adaptive region growing method. To establish road-prior knowledge database, a multivariate decision tree algorithm based on genetic programming was proposed. Though collecting a few training samples, prior knowledge database and road model are set up. Then the real road region can be detective according to the region growth and road model. Experiment results show that the algorithm is effectively, and is robust to complex environment such as illumination changes and shadow disturbances. 3) For real-time and robustness requirements of intelligent vehicle navigation, a multi-scale object-oriented stereo matching is proposed. By consideration of firmness, color and shape features, the rapid segmentation method is obtained. Then by employing a refined dynamic programming algorithm which is based on the inner edgeline of the segment, the disparity is obtained. According to the accuracy demand, determine whether need to do the refined segmentation. Finally, the final disparity images are gained. This multi-scale method is different from pyramid-layered method, which is more consistent with human vision process. The experimental results indicate that the algorithm has high matching precision and fast calculating speed.4) In view of the traditional path planning algorithm for intelligent vehicle ability to adapt to the complex poor environment, the artificial potential field method is improved. By improving potential field function, considering the obstacles and target movement tendency and using collision judgment constraint, the unnecessary collision in a dynamic environment is avoided.Introducing the target motion concept, the improved artificial potential field method is extended to target tracking function. Afterwards, the simulation results fully demonstrated the effectiveness of the method.