Research On The Application Of LiDAR In The Near-range Environment Perception Of Intelligent Vehicle

Intelligent vehicle is an important direction of automobile industry development.At present,the intelligent vehicle is still in the stage of research and development,among which there are problems such as low design speed,small operation design domain and so on.One of the important reasons for this problem is that the problem of environmental perception in the near-range of intelligent vehicle is not well solved.The existing product level automatic driving perception system is mainly equipped with millimeter wave radar and visual camera,which is difficult to realize the perception of small volume targets such as road shoulders,pedestrians and foreign objects in the near range.LiDAR has high measurement accuracy,which can be used to detect road shoulder and small obstacles.It is an important way to improve the design speed and application range of intelligent vehicle using LiDAR to perceive the near-range environment,but its technology application is not mature at present.The purpose of this project is to explore the application of LiDAR in the near-range traffic area of intelligent vehicles,and to make up for the shortage of millimeter-wave radar and visual camera in the near-range perception.The main research content includes the perception of road boundary in the form of shoulder and the accurate detection of obstacles,which is verified by the real vehicle experiment.In the aspect of road boundary awareness,the global path planning module of intelligent vehicle needs the data support of road boundary information.However,for the temporary changes of the road boundary,HD-map is often difficult to update and mark in time,and it is difficult to detect it with millimeter-wave radar and vision camera.This subject attempts to explore the use of LiDAR to solve the problem of road boundary detection in the near-range,mainly including three aspects: the external parameters of lidar are calibrated,and the LiDAR point cloud data is transformed into vehicle coordinate system;Based on the information fusion of LiDAR and GPS/INS,the global positioning of the main vehicle is realized,and the conversion relationship between the vehicle coordinate system and the global inertial coordinate system is established;In the vehicle coordinate system,the road boundary detection based on laser point cloud is realized by using the model fitting method,and the road boundary sensing information in the global inertial coordinate system is obtained by using the coordinate transformation relationship mentioned above.The experimental results show that the proposed scheme is effective and accurate.In the aspect of obstacle perception,the local behavior decision-making module of intelligent vehicle needs the data support of the state information of obstacles and traffic targets.However,the existing methods based on millimeter-wave radar and vision camera are difficult to realize the detection of small obstacles in the near-range road.This subject attempts to explore the use of highdefinition point cloud data of LiDAR to obtain the perception information of obstacles in nearrange roads,mainly including three aspects: using the point cloud data after the external parameter calibration based on the segmented plane fitting method to achieve the separation of road point cloud and non road point;Based on the classic DBSCAN density clustering method,the obstacle model in the road is detected.Here,in view of the problem of large computing and memory requirements of DBSCAN algorithm and the characteristics of near dense and far sparse target point cloud,this paper adopts the strategy of down sampling,KD-tree reorganizing and changing threshold;The nearest neighbor method is used to correlate the obstacle detection information of the front and back frames,and the motion state estimation of the obstacle relative to the vehicle coordinate system is realized based on the filtering theory.The experimental results show that the obstacle sensing method adopted in this paper has certain effectiveness and robustness for the detection of different volume obstacles.