LIDAR Ground Surface Reconstruction Using Virtual Grid Tiles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for reconstructing the ground surface topology using LIDAR data face challenges, particularly in non-flat environments and with non-calibrated sensors, as they require high computational load and struggle to accurately apply a-priori knowledge to sensor data.
Innovation Solution
A method employing a virtual horizontal grid to define stixels, using a tile fitting method and region growing to identify ground surface tiles, and recalculating height values, which allows for efficient reconstruction of ground surface topography even with tilted or out-of-calibration LIDAR sensors, avoiding the need for voxels and incorporating a-priori knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If voxel-based ground plane estimation is used, then ground surface reconstruction can be performed, but vertical resolution depends on voxel height requiring more voxels for higher resolution which increases computational load
Solution Approach 1:
The patent segments the point cloud data into ground points and non-ground points using a ground surface model that represents the ground surface as a collection of planar patches. This segmentation approach avoids the need for voxel-based representation, allowing high vertical resolution without increasing computational load through excessive voxel division.
Solution Approach 2:
The patent changes the fundamental parameter representation from 3D voxels to 2D ground surface patches with height information. By representing the ground surface as z = f(x, y) using planar patches instead of volumetric voxels, the system achieves high vertical resolution while maintaining lower computational complexity.
2Measurement precision
If a-priori knowledge is included in the detection algorithm, then ground surface reconstruction accuracy can be improved, but it is difficult to apply such a-priori knowledge correctly to the available sensor data
Solution Approach 1:
The patent incorporates a-priori knowledge by pre-defining the ground surface model structure as a collection of planar patches before processing the sensor data. This preliminary modeling approach guides the classification process and makes it easier to correctly apply domain knowledge about ground surface characteristics without increasing algorithmic complexity.
3Quantity of substance
If LIDAR measurements are taken from the reference point of the robot, then 3D environment measurement is achieved, but estimation of height at arbitrary points far from the robot becomes non-trivial in non-flat ground surfaces
Solution Approach 1:
The patent introduces a ground surface model as an intermediary between the LIDAR measurements and the height estimation task. This model, represented as z = f(x, y) with planar patches, serves as a mediator that connects the reference-point measurements to arbitrary point height estimates, making the estimation process straightforward even in non-flat terrains.
Data Source
AI summary
A method for reconstructing surface topology of a ground surface in an environment of a vehicle. The reconstruction is based on LIDAR sensor scan data. A virtual horizontal grid is defined for dividing the environment into distinct grid cells. In each grid cell, a local tile is defined as a function of coordinates of those measurement points that are contained in the respective grid cell. The respective local tile of each individual grid cell is classified as belonging to the ground surface based on reference surface data that identifies the respective local tile of at least one grid cell as a ground surface tile. For at least some grid cells, a respective height value of the measurement points in that grid cell is re-calculated as being the vertical distance of the respective measurement point above the ground surface, as it is defined by the identified ground surface tiles.