Research On Environmental Information Extraction And Movement Decision-making Method Of Unmanned Vehicle

With the growth of car ownership,the resulting road traffic accidents caused huge losses to the society and the people,among which the dangerous driving behavior of vehicle drivers is the main cause of frequent road traffic accidents.Because of its characteristic of no human manipulation,the unmanned vehicle has a broad application prospect and has been the attention of countries all over the world.In the course of the driving of unmanned vehicle,how to extract robust driving environmental information in real time and to make reasonable movement decision-making based on the obtained information are the keys to realize safe and efficient autonomous driving.Based on the national natural science foundation of major research project(90920305)“Research and Development on Intelligent Test Environment of Unmanned Vehicle” and the central university fund innovation team project(CHD2011TD006)“Research on Key Technologies of Unmanned Intelligent Vehicle Based on Visual Information”,the environmental information extraction and the movement decision-making method of unmanned vehicle are studied in this paper to achieve safe and efficient driving.The main research work of this paper is as follows:(1)Visual image data acquisition model and pretreatment research.With the unmanned vehicle coordinate system as the constraint condition,a visual image data acquisition model is established;and then,aiming at the problem that the quality of image acquisition is easily influenced by the driving environment,the multi-scale Retinex image enhancement algorithm and the traditional median filter algorithm are improved;finally,in order to verify the effectiveness of the improved algorithm,a static off-line contrast experiment is carried out.(2)Aiming at the problem of poor robustness of lane mark detection algorithm in complex road environment,an improved road image segmentation method for image pixels is proposed and the lane mark profile information is excavated in depth;and on this basis,with the combination of bidirectional scanning of sampling lines and candidate feature points with imaging model constraints,a lane mark detection optimization algorithm is proposed.In order to realize the effective switch of lane mark detection and tracking module,a confidence judgment module and a failure judgment module are established.(3)Aiming at the problem of the unbalance between efficiency and robustness of the unstructured road boundary detection,an unstructured road boundary detection optimization algorithm based on confidence probability block-based classification and improved least square method is proposed.Finally,a static off-line contrast experiment is conducted to verify the effectiveness of the proposed detection algorithm.(4)Aiming at the problem that the unmanned vehicle movement decision-making system requires high accuracy and stability of front vehicle identification and localization,a front vehicle identification algorithm based on vision sensor and 64-line 3D lidar information fusion is proposed.The region of interest(ROI)of the front vehicle in the image is determined by merging the obstacle position information extracted by the 3D lidar;the Haar-HOG fusion feature is used as the target vehicle description method,the AdaBoost algorithm is used for off-line vehicle training to obtain the cascade classifier for front vehicle identification,and the ROI of the unidentified front vehicle is further analyzed to solve the leakage identification problem caused by occlusion.(5)Research on movement decision-making modeling method of unmanned vehicle.Based on the premise of macroscopical driving,the two basic driving modes and six typical driving behaviors in the micro-dynamic traffic environment are further studied on the basis of the extraction of environmental information and the status of unmanned vehicle;the movement decision-making condition and the corresponding target quantity of the unmanned vehicle motion model are established.And based on this,the movement decision-making model based on the decision tree is established.Finally,the rationality is verified by constructing the micro-dynamic traffic simulation environment.(6)The unmanned vehicle platform is built based on the host computer units and the parameters of the generalized vision sensor system are calibrated,based on which a road experiment is carried out to verify the effectiveness of the related algorithm.