Research On Intelligent Vehicle Active Obstacle Avoidance Technology Based On The Fusion Of Millimeter Wave Radar And Lidar

With the rapid development of automotive industry,intelligent vehicles,which integrate environmental awareness,planning and decision-making,autonomous driving and other functions,have become a hotspot in the field of vehicle engineering.Active obstacle avoidance technology is an important research content to ensure the driving safety of intelligent vehicles.It can judge the potential collision danger in real time and avoid the collision by active braking or steering,which can effectively reduce the accident rate.This paper focuses on the active obstacle avoidance technology of intelligent vehicles based on millimeter-wave radar and Li DAR,and mainly carries out the following research works:(1)Based on the analysis of the characteristics of sensors and the functional requirements,the sensor configuration scheme of "millimeter-wave(MMW)radar & Li DAR" is selected to complete the environment sensing system.Therefore,the selection,installation and joint calibration of the sensors,together with the layered software architecture design,is completed to form the hardware platform and software environment support of this paper;(2)The data fusion method of MMW radar and Li DAR is explored.This part is start with sensor data preprocessing,where the near-field and far-field data fusion and relative velocity processing are carried out for MMW radar data,and Li DAR data is processed by angle interpolation and median filtering.On this basis,the data fusion method is analyzed,the data fusion structure and asynchronous data fusion algorithm are designed,and the accuracy of the fusion results is verified and improved through experiments and error analysis,which provides a more accurate raw data for the environmental awareness of intelligent vehicles;(3)Research on the algorithm of road area recognition to filter out irrelevant data points during vehicle driving.Firstly,a pass-through filter is used to delineate the region of interest,and then the ground points are segmented and filtered based on RANSAC algorithm.For the remaining data points,a road model is established for comprehensively analyzing the characteristics of road curbs,then a road curb recognition algorithm based on geometric features and three-dimensional point cloud features is designed.The extracted curb points are successively fitted along the curve by RANSAC algorithm and least square method,so as to obtain the points in road area.Experiments show that the algorithm is of good real-time performance,accuracy and robustness;(4)Combined with the characteristics of the fused point cloud of MMW radar and Li DAR,the problem of obstacle clustering in the road area is solved.Based on the analysis of the characteristics of the classic clustering algorithm,enhancements for the traditional DBSCAN algorithm are proposed,which mainly include dimensionality reduction at the input end,data hierarchically structured,changing radius,division of core area and changing the search method.Then effective targets are selected according to the characteristics such as size,height from the ground and so on.Experiments prove that the improved DBSCAN algorithm has good accuracy and real-time performance,and can provide real-time road environment information for intelligent vehicles;(5)The characteristics of typical multi-objective association algorithms are analyzed,which leads to the design of a multi-feature weighted nearest neighbor data association algorithm based on life cycle,and combined with Kalman filter,the association and tracking of multi-frame data are completed.On this basis,obstacle avoidance strategies in some typical scenarios are formulated,including keep,change,follow and stop driving.Considering the obstacle avoidance strategy and existing safety distance models,the safety distance determination rules under different working conditions are established,and the lane change trajectory model superposed by constant velocity offset and sine function is used to plan the lane change trajectory,which jointly advance the implementation of the obstacle avoidance strategy.Finally,the reliability of the data association tracking algorithm is verified through experiments,and the effectiveness of the active obstacle avoidance strategy is verified under typical scenarios.