Off-Road Path Selection via Lidar Depth Map Fusion

The patent introduces lidar as an intermediary sensor that creates depth maps and 3D representations of the off-road environment, serving as a mediator between the vehicle and the unstructured terrain. This allows the ADAS to navigate off-road by translating physical terrain into digital representations that can be processed and used for path planning, bridging the gap between traditional marker-based navigation and off-road conditions.

The system changes the fundamental parameters of environmental perception by switching from 2D image-based recognition (cameras) to 3D depth-based mapping (lidar). This parameter change enables the system to perceive and navigate complex off-road terrain with varying elevations, slopes, and obstacles that cannot be captured by traditional 2D visual markers alone.

The system performs preliminary scanning and mapping of the off-road environment using lidar before making navigation decisions. By creating depth maps and 3D representations in advance, the system prepares detailed environmental data that enables accurate path selection and safety assessments, allowing the vehicle to navigate complex terrain with precision even without pre-existing map data.

The patent replaces traditional mechanical navigation aids (road markers, physical signs) with optical sensing (lidar) and computational processing. This substitution enables the system to create its own navigation references from the environment itself, achieving both off-road adaptability and path selection accuracy through sensor-based environmental understanding rather than relying on pre-installed physical markers.

The patent segments the complex processing task into distinct stages: first processing lidar data to create depth maps, then processing camera images separately, and finally fusing these processed results for path determination. This segmentation of the computational workflow reduces overall complexity by breaking down the monolithic processing task into manageable, specialized sub-tasks that can be executed more efficiently.

The system performs preliminary processing of raw lidar point clouds into structured depth maps and preliminary image processing before fusion. By pre-processing and organizing the complex sensor data into standardized formats in advance, the system reduces the computational burden during the final path determination stage, achieving accurate results with more efficient processing.