Research on Fusion Localization Algorithm Based on Vision and LiDAR

The Apriltag algorithm has gained widespread recognition for its robustness and high precision in robot positioning on both domestic and international fronts. However, challenges persist in terms of real-time performance and accurate depth estimation, especially during high-speed motion and changing lighting conditions. To address these limitations, this paper focuses on refining obstacle extraction and tracking algorithms within point cloud models. Additionally, it introduces a novel solution – a two-dimensional code obstacle localization algorithm based on the fusion of visual and LiDAR data. By converting three-dimensional Apriltag localization coordinates into a three-dimensional point cloud format, multi-sensor data fusion is achieved. This results in more dependable and consistent relative positioning data, effectively overcoming issues of low real-time positioning accuracy caused by target motion. The algorithm is implemented within the ROS system, and its real-time performance is validated by comparing it with the Apriltag algorithm using the Gazebo simulation platform. Furthermore, feasibility is demonstrated by comparing positioning measurement data with real-world observations. In summary, this research pioneers an innovative approach to enhance the Apriltag algorithm through visual and LiDAR fusion. The algorithm’s real-time capability is substantiated through comprehensive simulations, with viability confirmed through robust validation against actual data. This advancement holds the potential to significantly improve robot positioning accuracy, even within complex and dynamic environments.