Low-Capacity 2D NDT Map for Autonomous Vehicle Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving systems face challenges in efficiently localizing vehicles due to the large data size of 3D point cloud maps, which limits the scalability of driving areas and increases computational requirements.
Innovation Solution
The method involves generating a low-capacity 2D normal distribution transform (NDT) map from a 3D point cloud for a predetermined region, allowing for efficient compression of map data and improved scalability of autonomous driving areas. This is achieved by gridding the 3D point cloud, modeling points as normal distributions, and storing this information in a 2D grid structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a 3D point cloud map is used for autonomous driving localization, then localization accuracy is improved, but map data size becomes too large limiting scalability
Solution Approach 1:
The patent transforms 3D point cloud data into 2D NDT map representation by projecting three-dimensional spatial information onto a two-dimensional grid structure. This dimensional reduction maintains essential localization features while dramatically reducing data storage requirements, enabling scalable autonomous driving systems for wide areas.
Solution Approach 2:
The patent changes the representation parameters from raw 3D point coordinates to probabilistic normal distribution parameters (mean, covariance) organized in 2D grids. This parameter transformation compresses large volumes of point cloud data into compact statistical representations that preserve localization accuracy while reducing data size.
2Measurement precision
If a 3D NDT map is generated from 3D point cloud data, then localization performance is improved, but computational load increases
Solution Approach 1:
The patent reduces computational complexity by operating in 2D space rather than 3D space. The 2D NDT map structure requires fewer calculations for probability density evaluation and map matching operations, significantly lowering the computational power needed while maintaining localization performance.
Solution Approach 2:
The patent simplifies computational operations by using 2D grid coordinates and associated normal distribution parameters instead of full 3D point cloud processing. This parameter reduction enables faster probability calculations and more efficient localization algorithms with lower computational overhead.
3Quantity of substance
If a 2D NDT map is generated for compression, then map data size is reduced improving scalability, but localization accuracy may deteriorate
Solution Approach 1:
The patent enhances local representation quality in the 2D NDT map by organizing normal distribution parameters in grid cells that capture local geometric features. Each grid cell contains statistically optimized parameters that preserve fine-grained localization information, ensuring that compression does not sacrifice accuracy in critical areas.
Solution Approach 2:
The patent performs preliminary statistical processing of 3D point cloud data during map generation, pre-computing normal distribution parameters for each 2D grid cell. This preliminary action extracts and stores essential localization features in compressed form, enabling accurate real-time localization without requiring access to original 3D point clouds.
Data Source
AI summary
Provided are a method, device, and recording medium for localizing an autonomous driving vehicle using a low-capacity NDT map. The method for localizing an autonomous driving vehicle using a low-capacity NDT map according to various embodiments of the present invention is executed by a computing device and may comprise the steps of: collecting 3D point clouds for a prescribed area; and localizing an autonomous driving vehicle using the collected 3D point clouds and a capacity normal distribution transform (NDT) map generated in correspondence with the prescribed area.