Adaptive LIDAR Depth Map Sampling for Small Object Detection

The patent segments the scene into multiple regions based on image analysis, identifying areas with objects versus empty spaces. The LIDAR scan is then applied selectively to only those segments containing objects, rather than uniformly scanning the entire scene. This segmentation approach maintains complete coverage of relevant areas while eliminating redundant scans of empty regions, thereby reducing scan time without compromising depth map quality.

The patent implements local quality by applying different scan densities to different regions of the scene. Areas containing objects receive higher scan density for accurate depth capture, while empty regions receive reduced or no scanning. This localized adaptation of scan quality based on scene content allows the system to maintain measurement precision where needed while minimizing time consumption in irrelevant areas.

The patent performs preliminary action by using a camera to capture an image of the scene before conducting LIDAR scanning. This pre-scan image analysis identifies the locations of objects and determines which regions require dense LIDAR sampling. By preparing this spatial map in advance, the system can then concentrate power-intensive dense sampling only where necessary, rather than uniformly across the entire scene, thus reducing overall power consumption while maintaining depth map accuracy in critical areas.

The patent applies partial action by performing dense LIDAR sampling only on portions of the scene that contain objects, rather than excessively sampling the entire scene. The system determines the minimum necessary sampling density required to capture object depth information accurately, avoiding excessive sampling in empty regions. This partial application of dense sampling maintains measurement precision where needed while significantly reducing power consumption.