LA-VIO: LiDAR-Assisted Visual-Inertial Odometry with Structural Feature Analysis

Line and vanishing points, as crucial geometric elements in visual scenes, provide additional structural information that can significantly enhance the accuracy of point-only based Visual-Inertial Odometry (VIO) systems. However, existing methods face challenges such as inaccurate line depth estimation and unreliable vanishing point extraction under varying lighting conditions and dynamic scenes. To address these issues, this paper proposes a method for line depth estimation and vanishing point extraction assisted by LiDAR point clouds, aiming to achieve more accurate and reliable spatial positioning and navigation in complex environments. The proposed method first acquires LiDAR points on the back-projection plane of image lines, thereby associating line depth with high-precision point cloud depth. Next, it extracts structural planes from the LiDAR point cloud and projects the normal vectors of these planes onto the image to assist in the extraction of line-based vanishing points. Finally, the method integrates line factors and vanishing point factors into a factor graph and employs a joint optimization algorithm to further enhance the accuracy of pose estimation. Extensive experiments conducted on the KITTI benchmark, M2DGR datasets, and real-world scenarios demonstrate the high accuracy and robustness of the proposed method in various environments. The experimental results confirm the effectiveness of integrating LiDAR point cloud data in improving the performance of point-only VIO systems.