Research On Automatic Driving Technology Based On Lidar

With the rapid development of sensing sensors,the use of lidar for environmental sensing has become one of the key research topics in the field of autonomous driving.Lidar is not easily affected by the weather and can directly obtain the basic shape,position,and intensity of obstacles.It is widely used in environmental perception and positioning of autonomous driving.This paper focuses on the research of motion compensation for lidar,ground detection in environmental perception,obstacle detection,tracking of obstacles,and positioning of lidar and inertial navigation.The main work of this article is as follows:1.Lidar motion compensation.Lidar motion compensation is an unavoidable process in the dynamic background target detection of smart cars.First,the quaternion method is used to solve the pose change matrix of the vehicle body in the previous scan period and the current scan period.Second,according to the characteristics of the static scene and the data packets generated by historical lidar data frames,the background in the time T coordinate system is modeled using a Gaussian mixture model to obtain the feature points of the moving target at time t,and the feature points are compared with the current frame.Matching points match to further refine the point’s new position in the current frame.2.Ground detection by lidar.The precise ground segmentation of Lidar point clouds is an important pre-processing task for autonomous driving environment perception,and is the basis for obstacle detection,classification and dynamic target tracking.First remove the invalid points before selecting the feature points.Then divide the circle centered by lidar into 180 sectors with 2°as a part,find the point with the smallest value projected on each sector,and use these points as a basis to filter out each sector and the area.The height difference between the minimum points matches the points in the range,and the plane equation is obtained by plane fitting using a random sampling consistency algorithm.Finally,loop through all the points,compare the values of these points with the height of the fitted plane,and perform ground detection.3.Lidar obstacle detection and tracking.Lidar scanning point clouds are fan-shaped.Before performing obstacle detection and tracking,the polarized grid method is used to project and cluster the scanned obstacles to achieve the classification of the obstacles.Then,the main direction of the classified object is determined based on Huff’s straight line detection.The main direction of the line is rotated around the origin to be parallel to the X axis.It traverses all points to find the maximum and minimum points,and draws parallel to the main coordinates.In the outer frame,the data with the outer frame is rotated back to the original position,and the three-dimensional block diagram is determined by the minimum and maximum values of points in the Z-axis direction,and obstacles are identified according to the volume of the three-dimensional block diagram.Finally,the Hungarian algorithm is used to find the optimal solution of the correlation matrix between the tracked object and the newly detected object to achieve obstacle tracking.4.Lidar and inertial navigation achieve positioning.The positioning application scenario of this article is to use laser radar and inertial navigation to achieve short-term accurate positioning when the positioning data becomes unusable in the short-term failure of GPS.Firstly,GNSS and inertial navigation combined navigation data and lidar were used to construct a point cloud map of the experimental area.Then during actual positioning,point cloud matching is performed using the data scanned by the lidar and the high-precision map constructed in advance to determine the specific location of the unmanned vehicle in the map.5.Experimental research.This article builds an experimental platform based on RS-Li DAR-16 lidar and GPS/IMU integrated navigation,and verifies lidar-based motion compensation,ground detection,obstacle detection and tracking experiments.The test site is the roads around Sinan Navigation Company,and the average speed of the car is between 10-15 km/h.The experimental results show that the motion compensation algorithm proposed in this paper has good motion compensation performance and can be used in real-time detection of moving targets;the ground detection algorithm can realize the classification of surrounding obstacles and the ground.Obstacle detection and tracking experiments based on ground detection can accurately detect obstacles and achieve stable tracking of vehicles;perform lidar and inertial navigation positioning experiments based on the established maps,and realize real-time positioning when GPS fails,The final positioning error is less than or equal to 0.165 m.