Multi-Layer Map Creation for Low-Cost Precise Vehicle Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional positioning solutions for autonomous driving face a trade-off between high precision, which is costly and limited to mature laser point cloud methods, and low-cost image-based methods that lack the necessary accuracy for autonomous driving applications.
Innovation Solution
A map creation method that combines image feature layers and local feature layers using a camera and laser radar data to generate a high-precision map, allowing for precise positioning of moving entities without the need for expensive laser radars, utilizing a camera and position sensor for real-time three-dimensional information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If laser radar is used for positioning, then positioning precision is improved, but cost increases
Solution Approach 1:
The patent merges image-based positioning and laser point cloud positioning into a fused positioning system. The image processing module extracts visual features while the laser point cloud module extracts spatial features, and their results are fused to achieve high positioning precision without relying solely on expensive laser radar hardware for every vehicle.
Solution Approach 2:
The patent creates a high-precision map in advance using laser radar data during the mapping phase. This pre-built map serves as a reference that can be used for positioning without requiring real-time laser radar measurements, allowing cost reduction while maintaining precision through the use of pre-acquired laser data.
2Ease of manufacture
If image-based positioning is used, then cost is reduced, but positioning precision deteriorates
Solution Approach 1:
The system combines image-based positioning with laser point cloud positioning by fusing features from both sources. The image processing provides visual context while the laser point cloud provides precise spatial information, and their fusion compensates for the limitations of image-only methods, achieving autonomous driving-level precision with lower cost.
Solution Approach 2:
The patent introduces a high-precision map as an intermediary that bridges image-based positioning and laser point cloud positioning. The map contains pre-extracted features from laser radar that serve as reference points, allowing the system to achieve high precision through image matching against this reference map without requiring real-time laser measurements.
3Measurement precision
If laser radar is deployed for autonomous driving, then positioning accuracy is improved, but device complexity increases
Solution Approach 1:
The positioning system is segmented into distinct modules: an image processing module that handles visual feature extraction, a laser point cloud module that handles spatial feature extraction, and a fusion module that combines their outputs. This segmentation allows each module to specialize in specific tasks, reducing overall system complexity while maintaining high accuracy through coordinated operation.
Solution Approach 2:
The patent uses a pre-built high-precision map as a reference copy that contains laser radar features. Instead of requiring every vehicle to perform complex real-time laser point cloud processing, the system matches images against this pre-processed map, significantly reducing computational complexity while maintaining positioning accuracy.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
Exemplary embodiments of the present disclosure provide a method, apparatus and computer readable storage medium for creating a map and positioning a moving entity. A method for creating a map includes acquiring an image acquired when an acquisition entity is moving and location data and point cloud data associated with the image, the location data indicating a location where the acquisition entity is located when the image is acquired, the point cloud data indicating three-dimensional information of the image. The method further includes generating a first element in a global feature layer of the map based on the image and the location data. The method further includes generating a second element in a local feature layer of the map based on the image and the point cloud data, the first element corresponding to the second element.