A High-Precision 3D Lidar Odometry Based on Image Semantic Constraints

Simultaneous localization and mapping (SLAM) are playing an increasingly important role in the field of robotics and autonomous driving. However, when mapping indoor and outdoor large-scale scenes, due to the sparseness of multi-line lidar point clouds, it is difficult to extract features in some scenes, which seriously affects the accuracy and robustness of lidar odometry. Aiming at this problem, this paper proposes an accurate localization algorithm for 3D lidar based on image semantic information constraints. Firstly, the lidar and camera timestamps are synchronized, then the point cloud is projected to the image coordinate system through the transformation matrix; Secondly, the static target semantic information is extracted through the SFSegNets network, and the depth value of the pixels in the area is obtained by interpolation according to the depth information of the point cloud, so as to realize the calculation of the semantic 3D position; Finally, the extracted semantic information of the image is used as landmarks to form constraints with the point clouds at the back end of the lidar odometry for localization optimization. In order to verify the performance of the algorithm proposed in this paper, experimental tests are carried out on the KITTI dataset and the actual campus scene, which both show that loop closures can be effectively detected after adding semantic information constraints. In the KITTI data set experiment with evo evaluation, the absolute error of the odometry has an average improvement of 1.64% after adding image semantic constraints.