Semantic HD Map Filtering for Obstruction-Free Vehicle Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in navigation due to the limitations of conventional maps, which lack the accuracy required for safe operation, as they often rely on sensor data that can be obscured or unreliable, leading to potential errors in vehicle localization and decision-making.
Innovation Solution
A system that stores high-definition (HD) map data with semantic labels, allowing for semantic label-based filtering to enhance visibility of critical objects, such as traffic signs, by excluding obstructing objects from view, thereby improving the accuracy of vehicle localization and navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional maps are used for autonomous vehicle navigation, then the system complexity is reduced, but the localization accuracy deteriorates to over 100m error
Solution Approach 1:
The patent segments the map data into multiple hierarchical levels: conventional map data for general navigation and HD map data for precise localization. The system selectively uses HD map segments only when high precision is required, maintaining low complexity for general operations while achieving high accuracy when needed.
Solution Approach 2:
The patent transitions from 2D conventional map representations to 3D HD map representations with detailed spatial information. This dimensional enhancement allows the system to achieve centimeter-level localization accuracy by utilizing vertical and depth information not present in traditional flat maps.
2Speed
If sensor data is used for autonomous vehicle operation, then real-time detection capability is improved, but reliability deteriorates due to obstructions and limited observation
Solution Approach 1:
The patent merges sensor data with HD map data to create a complementary navigation system. The HD map provides reliable spatial context and object information that compensates for sensor limitations, while sensors provide real-time updates. This combination maintains fast detection response while significantly improving reliability through multiple data sources.
Solution Approach 2:
The HD map acts as an intermediary that fills gaps in sensor observation. When sensors cannot detect objects due to obstructions or limitations, the HD map provides predictive information about expected objects and their locations, maintaining reliable operation without requiring direct sensor detection of every element.
3Measurement precision
If HD map data with semantic labels is used, then localization accuracy is improved to centimeter level, but device complexity increases
Solution Approach 1:
The patent implements dynamic map data loading and processing, activating HD map semantic label processing only when and where high precision localization is required. The system adapts its complexity level based on operational context, using detailed semantic analysis selectively rather than continuously, thus achieving high accuracy without permanent high complexity.
Solution Approach 2:
The patent applies semantic label processing and high-detail HD map analysis locally to specific regions and situations rather than globally. The system focuses computational resources on areas requiring precise localization while using simpler processing elsewhere, optimizing the balance between accuracy and complexity through spatially differentiated processing quality.
Data Source
AI summary
The autonomous vehicle generates an overlapped image by overlaying HD map data over sensor data and rendering the overlaid images. The visualization process is repeated as the vehicle drives along the route. The visualization may be displayed on a screen within the vehicle or at a remote device. The system performs reverse rendering of a scene based on map data from a selected point. For each line of sight originating at the selected point, the system identifies the farthest object in the map data. Accordingly, the system eliminates objects obstructing the view of the farthest objects in the HD map as viewed from the selected point. The system further allows filtering of objects using filtering criteria based on semantic labels. The system generates a view from the selected point such that 3D objects matching the filtering criteria are eliminated from the view.