3D LiDAR Voxel Filtering for Material Handling Vehicle Obstacle Detection

The patent divides the 3D space into discrete voxel cells, organizing the point cloud data from LiDAR sensors into a grid structure. Each voxel cell independently processes and filters data points within its boundaries, enabling parallel processing and reducing the computational burden on the entire system while maintaining comprehensive obstacle detection coverage.

The patent extracts and removes redundant data points from the point cloud by applying density-based filtering within each voxel cell. Cells with data point density below a threshold are marked as unoccupied and their data points are discarded, while only occupied cells with sufficient data density are retained, significantly reducing the total data quantity while preserving essential obstacle information.

The patent segments the point cloud processing task into independent voxel cell operations. Each cell performs local density analysis and filtering independently, allowing parallel processing across multiple cells. This segmentation transforms a single large computational task into many small parallel tasks, dramatically improving processing speed while maintaining detection reliability through comprehensive coverage of all cells.

The patent applies partial processing by selectively retaining only occupied voxels that meet the density threshold criteria, discarding unoccupied voxels without full processing. This partial action approach processes only the necessary portion of data (occupied cells) while skipping redundant portions (unoccupied cells), thereby improving productivity without compromising the reliability of actual obstacle detection.

The patent implements dynamic thresholding where the data point density threshold is adjusted based on the distance of each voxel cell from the LiDAR sensor. Cells at different distances have different threshold values, with farther cells using lower thresholds to account for beam divergence and reduced point density at distance. This dynamic adaptation maintains detection accuracy across varying distances while preserving efficient processing through automated threshold selection.

The patent applies different filtering criteria (threshold values) to different spatial locations (voxel cells at different distances from the sensor). Each cell receives a locally optimized threshold based on its distance from the LiDAR, rather than applying a single global threshold. This local quality approach ensures that each region is processed with appropriate sensitivity, improving overall detection accuracy while maintaining processing efficiency through localized rather than universal processing rules.