Map Matching Using Vector Semantic Features for Vehicle Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle positioning technologies, such as vSLAM and lSLAM, require large storage resources and additional descriptors, leading to inefficiencies and reduced accuracy in real-time matching, especially with sparse vector semantic maps.
Innovation Solution
A map matching method that determines observation semantic features from vehicle sensor data and matches them with map semantic features using vector information, without relying on additional descriptors or intensity, to achieve real-time matching and reduce storage and computing power requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If vSLAM or lSLAM technology is used for vehicle positioning, then positioning accuracy is improved, but storage resources are heavily occupied due to dense positioning maps and stored descriptors
Solution Approach 1:
The patent extracts only the essential vector semantic information from the dense positioning map, storing only key map features (lane lines, intersections, landmarks) in vector format rather than storing complete dense map data with all descriptors. This extraction approach maintains positioning accuracy while dramatically reducing storage requirements.
Solution Approach 2:
Instead of using dense maps and inverting the approach, the patent uses sparse vector semantic maps directly for positioning. The system inverts the traditional approach by building positioning capability on top of minimal vector map data rather than on top of comprehensive dense maps, achieving both reduced storage and maintained accuracy.
2Quantity of substance
If vector semantic map is used to reduce map size, then storage requirements are reduced, but matching accuracy deteriorates due to lack of descriptors and intensity information
Solution Approach 1:
The patent changes the parameter representation from traditional descriptor-based features to vector semantic features. By transforming map features into vector representations with semantic meanings (lane lines, intersections, landmarks) and using parameter changes in feature extraction and matching algorithms, the system achieves accurate matching with sparse vector map data without requiring traditional descriptors or intensity information.
Solution Approach 2:
The patent applies local quality by enhancing specific critical map features (intersections, landmarks, lane line configurations) with detailed vector semantic information at those locations, while keeping other areas sparse. This localized detail enhancement maintains matching accuracy at critical positioning points while preserving overall storage efficiency.
3Measurement precision
If dense positioning map with descriptors is used, then matching accuracy is improved, but real-time matching efficiency deteriorates due to computational complexity
Solution Approach 1:
The patent extracts only the essential vector semantic features needed for matching (lane lines, intersections, landmarks) from the dense map, removing unnecessary descriptors and intensity information. This extraction creates a streamlined vector semantic map that enables fast real-time matching while maintaining accuracy on critical features.
Solution Approach 2:
The patent segments the positioning map into discrete vector semantic elements (lane lines, intersections, landmarks) that can be independently processed and matched. This segmentation allows the system to efficiently search and match specific feature types without processing entire dense map data, improving real-time matching efficiency.
4Quantity of substance
If sparse vector map is used, then storage size is reduced, but device complexity increases due to special sensor requirements for additional descriptors
Solution Approach 1:
The patent creates a universal vector semantic map system that works with multiple sensor types (cameras, laser radars, millimeter-wave radars) without requiring special sensors or additional descriptors. The vector semantic features serve multiple functions across different sensor modalities, enabling the same sparse map structure to support various sensing approaches and maintaining system simplicity.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
Provided are a map matching method and apparatus, an electronic device and a storage medium. the method comprises: acquiring initial positioning information of a vehicle and vehicle sensor data (601); determining a plurality of pieces of candidate positioning information on the basis of the initial positioning information (602); determining a plurality of observation semantic features on the basis of the vehicle sensor data (603); acquiring local map information on the basis of the initial positioning information of the vehicle, the local map information comprising a plurality of map semantic features (604); for each piece of candidate positioning information (605): converting the plurality of map semantic features into a coordinate system of the vehicle sensor on the basis of the candidate positioning information, so as to obtain a plurality of candidate map semantic features under the coordinate system (6051); and matching the plurality of observation semantic features with the plurality of candidate map semantic features, so as to obtain matching pairs (6052); and determining optimal candidate positioning information of the vehicle and a matching pair corresponding to the optimal candidate positioning information on the basis of matching pairs corresponding to each piece of candidate positioning information (606). The method is suitable for a vector semantic map, achieves real-time matching of observation features and map features by only using vector information of the vector semantic map, does not depend on additional data such as additional descriptors and intensity, and achieves a good matching effect on the basis of reducing storage requirements and computing power use.