Research On High-precision Map Generation Method Based On Multi-sensor Fusion

With the advancement of technology,the development of smart cars has become more and more mature.Autonomous vehicles have gradually appeared in people’s vision in recent years.Auto-driving technology not only can decrease the occurrence of traffic malfunction,but also can improve travel efficiency,reduce the traffic jam.High-precision map technology is a vital part of the auxiliary automatic driving system,and it is also a key step for the real implementation of automatic driving.It is of great significance to study how to use sensors to establish a high-precision map of the road environment.This paper studies the high-precision map building method and proposes a mapping method based on information fusion.The main research contents are as follows:First of all,we build an acquisition system that meets the requirements of automatic driving based on electric wheelchairs,and complete the selection of sensors.The sensors involved include lidar,panoramic camera,combined IMU,odometer,etc.Considering the characteristics of each sensor and the angle of view,we complete the sensor layout design.Secondly,based on the point cloud data collected by lidar,the laser SLAM technology is used for simultaneous positioning and mapping,the campus three-dimensional point cloud map is established based on the ICP and NDT matching algorithms,and the Cartographer algorithm is used to perceive the surrounding environment information to establish high-precision three-dimensional points cloud map.Then,the panoramic camera is calibrated,based on the image data collected by the panoramic camera,the data is preprocessed,and different detection methods are used to realize the lane line detection of the road scene.For straight lanes,threshold masking and Hough transform methods are used to complete the detection;for curved lanes,a detection method based on sliding window polynomial fitting is used.Use deep learning algorithms to realize the detection and recognition of traffic signs in road scenes.For the speed limit traffic signs in the school,a convolutional neural network is built to train the traffic sign detection and recognition model,and the Alex Net network is used to recognize them,and the Faster-RCNN network is used to check them out.Finally,the campus road data is collected based on the unmanned electric wheelchair,and joint calibration of lidar and panoramic cameras is accomplished using the Autoware framework to fuse two-dimensional images with three-dimensional point clouds.In order to overcome the shortage of pavement information in single point cloud map and the limitation of low accuracy in single image vector map,this paper presents a high accuracy map making method based on multi-sensor information fusion,which projects vector map based on image onto high accuracy three-dimensional point cloud map based on point cloud.The final result map not only has road surface environment semantic information,but also with the high accuracy.