Recognition Method Research Of Traffic Markings On Auxiliary Navigation Of Unmanned Vehicle

Unmanned ground vehicles (UGV) is a smart car experimental platform which integrates various modules such as perception, decision and control and involves the knowledge in many subjects and technologies, such as phoronomics, kinetics, transducer, automatic control, artificial intelligence and pattern recognition. It synthesizes computer technology and robot technology and reflects the development and latest achievements of UGV. The ordinary automobile test can’t meet related performance requirements of UGV. However, the development and research of UGV project is a very complex project and need a mass of experiments with software and hardware development and debugging.Aware system is a very important part of UGV, which provides external environment information for UGV and collects the ambient conditions of the road by a variety of transducers including SICK radar, CCD vidicon and HLD, etc. Among them CCD vidicon is the main experimental instrument in this thesis and provides a variety of environmental sensory information.Real-time traffic marking identification system is the most important part of aware system of UGV which exerts important influences for its overall performance. Therefore, it has attracted more and more attention of researchers. However, because of the complex situation in natural conditions and high demands for the real-time system, the experiment results can not reach the ideal effects and many problems in this aspect are still remained to be solved.In this thesis, to begin with, we extract each frame from the video obtained by CCD mounted on UGV and deal with the traffic marking based on the images. Based on research achievement at home and abroad, we design the basic framework of the traffic marking identification system and explain in detail four key technologies: preprocessing technology in chapter2, edge detection technology in chapter3, feature extraction technology base on moment invariant in chapter4and template recognition technology in chapter5.Chapter4and5is the core part in this thesis, which proposes feature extraction and identification methods for traffic markings. Chapter4elaborates feature extraction methods for traffic markings base on moment invariant theory in detail. The moment invariant theory can successfully solve the problem of deformation markings in natural conditions, which remains the constant characteristics of rotation, translation and scale. Chapter5states the identification methods for traffic markings using template matching. Because of the large deformation or distortion of traffic markings in vehicle-mounted video images, the multi template matching is employed to increase recognition accuracy, which means each image of traffic marking has two or more templates. The traffic marking is recognized if it accords with one of the templates. This method can increase the identification accuracy but lengthen the processing time and we make a better balance between them.The results of the simulation experiments and actual tests in physical environment demonstrate the traffic marking identification system which has some practical applicability with faster processing speed and higher accuracy. It can meet the need of the real-time identification to a certain extent and provide essential navigation information for UGV.