Lane Detection Using Trajectory-Based Template Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing lane detection systems face challenges in accurately determining lane information when high-definition maps are unavailable or unreliable, such as due to GPS malfunctions, map updates, incorrect lane features, or insufficient image data, leading to unstable autonomous vehicle operations.
Innovation Solution
A method and device for lane detection that combines trajectory information from dynamic and static objects using a camera and radar data to generate adjusted lane templates, determining the best match through overlapping features, ensuring robust lane identification even in challenging conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-definition maps are used for lane detection, then lane recognition accuracy is improved, but system reliability deteriorates when maps are unavailable or incorrect
Solution Approach 1:
The patent introduces trajectory information from dynamic objects and static environment as intermediary data sources. When high-definition maps are unavailable or unreliable, the system uses radar and camera data to track trajectories of surrounding objects and environmental features, which serve as mediators to infer lane information without direct reliance on map data.
Solution Approach 2:
The system creates virtual lane templates by copying and adapting lane geometry from high-definition maps when available. These templates are then adjusted using real-time trajectory information from sensors, creating a hybrid representation that combines map-based prior knowledge with current environmental observations to maintain reliability when maps are incomplete or incorrect.
2Reliability
If multiple sensors and processing steps are used to improve lane detection reliability, then system reliability is improved, but device complexity increases
Solution Approach 1:
The patent merges trajectory information from multiple sources (dynamic objects via radar, static environment via camera) into a unified lane detection framework. By combining these different data streams and processing them through integrated algorithms, the system achieves reliable lane detection without requiring each individual sensor or processing step to be overly complex.
Data Source
AI summary
A lane detection method and device is disclosed, where the method includes determining main trajectory information based on first trajectory information of one or more dynamic objects around a vehicle and second trajectory information of a surrounding environment of the vehicle, determining one or more candidate lane templates from among lane templates based on the main trajectory information, adjusting the one or more candidate lane templates to generate adjusted lane templates, determining an overlap between a lane feature obtained from an input image and each of the adjusted lane templates through overlapping the lane feature with each of the adjusted lane templates, and determining, as the lane, an overlapping lane template in which an overlap of the lane feature with an adjusted lane template has a greatest similarity.