Vehicle Landmark Detection for Self-Learning Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current positioning systems for vehicles, especially partially automatically controlled ones, face challenges in utilizing available environmental information comprehensively for precise self-localization, often requiring complex software updates and specific detectors for various landmark types.
Innovation Solution
A method that generates and updates detector modules based on training data from environmental and map data, allowing for the detection of landmarks using machine learning, enabling the creation of up-to-date detector modules without complex software updates and facilitating the detection of various landmark types using existing vehicle sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex software updates are performed to update detector modules, then detection accuracy is improved, but system complexity and update time increase
Solution Approach 1:
The system performs self-learning by automatically generating updated detector modules from newly collected environmental data and map data without requiring external software updates. The learning unit continuously trains detector modules using machine learning algorithms, enabling the system to improve its own detection capabilities autonomously.
Solution Approach 2:
The system collects and processes environmental data in advance to generate training data before detection is needed. By pre-training detector modules using accumulated data from multiple sources (cameras, LiDAR, map data), the system prepares detection capabilities proactively rather than reactively updating software.
2Measurement precision
If specific detectors are created for each landmark type, then detection precision is improved, but device complexity increases
Solution Approach 1:
The learning unit serves multiple functions by collecting environmental data, generating training data, training detector modules, and updating the landmark database using the same processed data. This multi-functional approach reduces the need for separate specialized components for each landmark type while maintaining high detection precision.
Solution Approach 2:
The system merges data from multiple sensors (cameras, LiDAR, map data) and multiple processing functions into a unified machine learning framework. By combining these elements into an integrated learning system, the patent reduces overall system complexity while improving detection capabilities across all landmark types.
3Adaptability or versatility
If comprehensive environmental data is collected for all landmark types, then adaptability is improved, but data processing time increases
Solution Approach 1:
The system collects environmental data continuously and comprehensively from all available sensors, gathering more data than immediately necessary. This excessive data collection ensures adaptability to all possible landmark types and environmental conditions, while the machine learning process efficiently processes only the relevant portions when needed.
Solution Approach 2:
Environmental data is collected and stored in advance in the database before detection scenarios arise. By pre-collecting and archiving comprehensive environmental data from multiple sensors and time periods, the system prepares adaptability resources beforehand, reducing real-time processing requirements.
Data Source
Figure 1~2
Figure 3
AI summary
In the method of the invention, map data is provided that includes position information for first landmarks (24) of a first landmark type, environmental data is acquired, and the position of the mobile unit (1) is determined. Based on the position of the mobile unit (1), the acquired environmental data, and the position information, training data is generated and stored for the first landmarks (24). Based on the training data, a first detector module for detecting the first landmark type is generated. The positioning system of the invention includes a storage unit (2) for providing map data that includes position information for first landmarks (24) of a first landmark type.It further comprises a data acquisition unit (3) for acquiring environmental data, a localization unit (6) for determining the position of the mobile unit (1), and a processing unit (7) for generating and storing training data based on the position of the mobile unit (1), the acquired environmental data, and the position information for the first landmarks (24). It also comprises a training unit (8) for generating an initial detection module for detecting the first landmark type based on the training data.