EA2D-LSLAM: Environment Analysis-Based Adaptive Downsampling for Point Clouds in LiDAR SLAM

Efficient and accurate point cloud downsampling is crucial for large-scale 3D mapping and localization, especially in urban environments where irregular structures present significant challenges. Traditional downsampling methods, including voxel-based filtering and uniform sampling, often fail to consider the varying importance of different regions in the point cloud, leading to the loss of critical information or bad matches. This paper proposes a novel environment analysis-based adaptive downsampling method to address these limitations. The method utilizes Hessian matrix decomposition derived from the Iterative Closest Point (ICP) algorithm to identify the feature directions in the point cloud. The information matrix for each voxel is then calculated to assess the contribution of each point to pose estimation. The localizability in different directions is evaluated, and the point cloud is adaptively sampled based on its relevance to the pose constraints by projecting this matrix onto the feature vectors. This approach improves computational efficiency while preserving critical environmental features, particularly in irregular urban environments. Experimental results demonstrate the effectiveness of our method in preserving key features and reducing redundant data, offering an improved solution for point cloud downsampling in complex environments.