Research On SLAM For Driverless Cars Based On Lidar And Camera Fusion

In recent years,the driverless technology in the automotive field has been rapidly promoted due to the breakthrough of computer technology.As an important branch in the automotive field,driverless technology plays a very important role in human-computer interaction,active safety and other aspects.The vehicle unmanned driving system is mainly composed of three parts,including perception,decision and control.The perception part is one of the core of the whole system,and the vehicle positioning technology plays a pivotal role in the perception link.Global Positioning System(GPS)is the common positioning method of vehicles,but it is easy to lose signals under viaducts,on both sides of tall buildings,underground parking lots and other scenarios.In order to effectively solve problems of this kind of unmanned vehicle,carry on its own importance is given to complete the positioning technology of sensor,such as: the application of the SLAM technology.Therefore,in this paper,the SLAM technology of driverless vehicles is deeply studied,and the driverless platform is built on the basis of traditional electric vehicles,and the environment map is established on this platform and the positioning effect of the algorithm in this paper is investigated.Based on the in-depth study of visual mapping and laser mapping,this paper uses the point cloud data of lidar to assist camera mapping.The main research work is as follows:1.The imaging model of the camera is deduced and explained.The self-made black and white checkered grid is used in the camera calibration,and the joint calibration of camera and lidar is completed based on the camera calibration.2.Improved the threshold setting method of ORB-SLAM2 algorithm when extracting feature points.The threshold used in the original algorithm needs to be adjusted manually in different scenes,but this paper introduced the local adaptive threshold method when setting the threshold.Firstly,the image grid is segmented.Then the image contrast value of each grid is calculated,and the local threshold is set based on this value,and the feature points are extracted from each independent grid.Finally,the pixel coordinates of the feature points in the grid in the whole image are restored and stored in a quadtree structure.In the end,in the image data quantity more and more stable feature points.3.In the odometer,the scale estimation of feature points was completed by using the fusion point cloud data,and the reprojection error minimization model was built in the camera motion estimation to optimize the projection position error of 3D points.Pn P graph optimization was used in such problems as camera pose optimization,and BA cost function was constructed and pose graph optimization was used in global optimization.Using the word bag model completed two image description of similarity degree,finally further illustrates the key frame selection problem in the loopback detection.4.The test platform was built on the basis of the electric vehicle,and the point cloud data of the lidar,the image data of the camera and the high-precision GPS position information used as the truth value of the trajectory were collected in a variety of test scenes on campus.After the mapping and positioning work is completed,the positioning effect of the improved algorithm is investigated by using the high-precision GPS position information.