Non-Semantic Reference Data for Motor Vehicle Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining a motor vehicle's position, such as landmark-based localization and satellite navigation, face limitations in accuracy and availability, especially in regions without recognizable landmarks or with low-cost satellite receivers.
Innovation Solution
A computer-implemented method generates non-semantic reference data by clustering raw data points from environmental sensors using descriptors that characterize the surroundings, assigning a characteristic number based on informational gain for positioning, allowing for precise localization independent of semantic landmarks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If landmark-based localization is used, then positioning can be performed using semantic structures, but the number of available features is restricted and localization cannot be carried out in regions without landmarks
Solution Approach 1:
The patent extracts and removes the dependency on semantic structures and predefined landmark classes. Instead of requiring features to be categorized as specific landmark types (traffic signs, building edges), the system uses raw sensor data points and their geometric relationships directly, eliminating the restriction that only semantically classifiable features can be used for positioning.
Solution Approach 2:
The patent changes the parameter of feature representation from semantic categories to geometric and spatial parameters. By using descriptors based on geometric properties (distances, angles, relative positions) rather than semantic labels, the system can utilize any detectable feature in the environment, significantly increasing adaptability to different regions regardless of whether predefined landmarks are present.
2Measurement precision
If highly precise satellite receivers are used, then positioning accuracy is sufficient for highly automated or autonomous driving, but the expense is significant
Solution Approach 1:
The patent segments the positioning task into two parts: using inexpensive satellite receivers for coarse position estimation and using local geometric feature matching for fine position refinement. This segmentation allows the system to achieve high precision without requiring an expensive high-precision satellite receiver throughout, reducing overall cost while maintaining accuracy.
Solution Approach 2:
The patent introduces geometric feature descriptors and point cluster matching as an intermediary between the low-cost satellite receiver and the final high-precision position determination. The satellite receiver provides an initial position estimate, and the intermediary geometric feature matching process refines this to achieve the required precision for autonomous driving, avoiding the need for expensive direct high-precision satellite hardware.
3Measurement precision
If only semantic landmarks are used for localization, then positioning can be performed in regions with recognized features, but the number of available features is limited
Solution Approach 1:
The patent creates a universal feature representation system that can handle both traditional semantic landmarks and non-semantic geometric features through the same processing pipeline. The point cluster descriptors and geometric relationship analysis can work with any environmental feature, making the system multi-functional and able to utilize a much larger quantity of features from diverse sources (sensor data points, natural features, artificial structures) without requiring separate processing methods.
Data Source
AI summary
A computer-implemented method for generating non-semantic reference data for determining the position of a motor vehicle involves a set of raw data points being provided that models a stipulated surrounding area. A stipulated descriptor characterizing a property of the surrounding area is determined for each of the raw data points. At least one point cluster is generated by grouping the raw data points on the basis of their descriptors. A characteristic number relating to an information gain for determining the position of the motor vehicle is assigned to a first point cluster on the basis of the descriptors of the raw data points. The characteristic number is taken as a basis for storing feature information of the first point cluster on a memory unit as non-semantic reference data for determining the position.