HD Map Refinement via Aerial Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for creating high definition (HD) maps of parking areas for autonomous valet parking are time-consuming, expensive, and fail to capture dynamic details like parked cars and thin obstructions, limiting the deployment of autonomous valet parking features due to the need for extensive vehicle-driven data collection and post-processing.
Innovation Solution
The system uses feature extraction from high-resolution satellite or drone images to identify parking spots, boundaries, and obstructions, allowing for the automated generation and refinement of HD maps without requiring vehicles to drive through the area, employing a hierarchical framework for image analysis and semantic grouping to create accurate and detailed maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If vehicle-driven data collection and post-processing methods are used to create HD maps, then map detail accuracy is improved, but time consumption and cost increase significantly
Solution Approach 1:
The patent uses aerial images as a copy or representation of the parking area to extract map features, eliminating the need for actual vehicle-driven data collection. Image processing algorithms analyze the aerial images to identify parking spaces, boundaries, and obstacles, creating HD maps without physical vehicle traversal.
Solution Approach 2:
The patent replaces the mechanical system of vehicle-driven data collection with an optical/image-based system. Instead of vehicles physically driving through parking areas to collect sensor data, the system uses aerial imaging and computer vision algorithms to extract map information, significantly reducing time and resource requirements.
2Loss of information
If vehicle-driven data collection methods are used to create HD maps, then map completeness is improved, but cost increases significantly
Solution Approach 1:
The patent uses aerial images as a cost-effective copy of the parking area environment, capturing comprehensive spatial information without the high costs associated with deploying fleets of vehicles equipped with sensors for data collection.
Solution Approach 2:
The aerial imaging system serves multiple functions: it captures parking space locations, identifies boundaries, detects obstacles, and provides overall area coverage, replacing multiple specialized vehicle sensor systems with a single multi-functional imaging approach.
3Stability of the object's composition
If traditional map generation methods are used, then static boundaries are captured, but dynamic details like parked cars and thin obstructions are missed
Solution Approach 1:
The patent employs dynamic image processing algorithms that can detect and classify various objects including parked cars, thin obstructions, and other dynamic elements in the parking area. The system analyzes image data to identify not only static boundaries but also temporary or movable objects that affect vehicle navigation.
Solution Approach 2:
The patent applies different analysis techniques to different regions and features in the aerial images. Specific algorithms are used to detect thin obstructions in certain areas, while other methods identify parked cars in parking spaces, ensuring that local features with different characteristics are captured with appropriate detail levels.
Data Source
AI summary
The present disclosure relates to refining maps for vehicles, for example using feature extraction to identify parking areas and their contents from high resolution satellite or drone images. Such maps can be useful for autonomous valet parking (AVP) features of autonomous vehicles.