Location Algorithm Of Unmanned Vehicle Based On Point Line Feature And Vision Laser Fusion

With the continuous progress of science and technology in recent years,driverless technology has been widely concerned by the industry,and the positioning of unmanned vehicles is an extremely important part.However,the visual SLAM algorithm based on feature points is easy to fail due to the lack of enough feature points in low texture scenes,At the same time,the visual features with rich image texture information and the laser features with high-precision environmental geometric structure information are naturally complementary.Based on this,this paper proposes a vision laser fusion algorithm for unmanned vehicle location based on point and line features.By adding the constraint of line features to enhance the robustness and accuracy of the system,the motion pose estimation of unmanned vehicle and the construction of point cloud map of surrounding environment are realized.The main research work of this paper is as follows:(1)The vision and laser data preprocessing module is constructed.For the visual feature points,Shi Tomasi corner extraction algorithm is used,and Lucas Kanade optical flow algorithm is used to track the inter frame feature points based on the consideration of real-time.LSD feature extraction algorithm is adopted for visual line features,and the improved algorithm stitches the line features with close distance and similar expression.In order to avoid the feature distribution is too concentrated,the minimum interval threshold is set to ensure the uniform distribution.For laser point cloud,firstly,based on the assumption of uniform motion,the motion distortion of point cloud is removed by linear interpolation method.Then,the molecular region is delimited and whether it is edge or plane feature point is judged according to the local curvature of point cloud.Finally,the invalid feature points such as occlusion and nearly parallel to the object plane are removed to enhance the effectiveness of the system.(2)Visual feature depth recovery module.The data association is realized by calibrating the external parameters of the sensor and aligning the data time stamp.Next,the laser neighborhood is selected to fit the plane,and the depth value of the intersection of the ray formed by the camera optical center and the visual feature point and the fitted plane is used to restore the depth of the visual feature point.(3)Vision laser tightly coupled odometer module.Firstly,the visual features and point cloud features are matched between frames,and the epipolar constraints are constructed to solve the initial pose estimation.Next,the loss function of visual features and the loss function of laser features are put into the same optimization framework for LM nonlinear optimization in order to solve the motion pose estimation.Finally,the point cloud environment map is constructed by laser mapping.(4)Based on Kitti data set and campus field experiment.Finally,this paper carries out the experiment of vision laser fusion unmanned vehicle positioning algorithm,based on the open source Kitti data set and the loam algorithm,the results show that the output accuracy of the odometer module proposed in this paper is higher;Finally,the effectiveness of the algorithm is tested through the collected campus data set of Harbin Institute of technology,and the algorithm runs successfully.