Vehicle Localization Using Stereo Camera Matching Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Satellite navigation in urban areas is prone to errors due to interference from obstacles, and existing correction technologies like DGPS and RTK are limited by the high cost and availability of sensors required for dead reckoning, leading to shadow areas where accurate vehicle localization is impossible.
Innovation Solution
A precise vehicle localization apparatus and method using stereo cameras to extract matching vectors from images, combined with GPS information and map data, for accurate vehicle localization, which includes a region of interest setting, edge image conversion, line component detection, image fusion, and dominant depth value extraction to estimate vehicle positioning coordinates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensors for dead reckoning are loaded into each vehicle to improve localization precision, then measurement precision is improved, but device cost increases significantly
Solution Approach 1:
The patent creates a virtual map by copying and storing visual features (landmarks, edges, lines) from the real environment into a digital representation. This virtual map serves as a reference for localization without requiring expensive sensors in each vehicle. The system copies environmental information into a usable format that can be matched against current camera images to determine position.
Solution Approach 2:
The patent replaces the mechanical sensor-based dead reckoning system with an optical-computational system using stereo cameras and image processing. Instead of using mechanical sensors to directly measure position and orientation, the system uses visual feature extraction, matching, and computational geometry to infer vehicle location and attitude from camera images and virtual map comparison.
2Measurement precision
If base station correction data is used to improve satellite navigation accuracy, then measurement precision is improved, but shadow areas still exist where signals cannot be received
Solution Approach 1:
The patent introduces a virtual map as an intermediary reference system that does not rely on satellite signals. Instead of directly using satellite corrections that can be blocked, the system creates a mediator (virtual map with visual features) that can be continuously referenced by stereo cameras regardless of satellite visibility, enabling localization in shadow areas where GPS signals are blocked.
3Ease of manufacture
If stereo camera image processing is used to extract matching vectors for localization, then device cost is reduced, but device complexity increases due to multiple processing steps
Solution Approach 1:
The patent segments the complex image processing task into distinct functional modules: feature extraction (identifying landmarks, edges, lines), virtual map construction (storing processed features with position data), and matching (comparing current features with virtual map features). This segmentation allows each module to be optimized independently and facilitates implementation using standard computer vision techniques.
Solution Approach 2:
The patent performs preliminary processing to construct a virtual map in advance by extracting and storing visual features from the environment along with their corresponding position information. This pre-processed virtual map serves as a ready-to-use reference, eliminating the need for complex real-time processing during localization. The system prepares the reference data beforehand, simplifying the real-time matching operation.
Data Source
AI summary
A vehicle localization method includes obtaining a stereo image while a vehicle is driven, extracting a matching vector from the stereo image while the vehicle is driven, loading a map vector of a current location based on GPS information in previously constructed map data, matching the extracted matching vector with the map vector, and estimating a positioning coordinate corresponding to the extracted matching vector as a current location of the vehicle when the extracted matching vector and the map vector meet a condition.