HD Map Generation Using Bird's-Eye-View Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous driving systems face challenges in generating high-definition maps using data from multiple sources, such as cameras and LiDAR, due to the difficulty in effectively fusing and aligning features from different data types, leading to inaccuracies and reduced robustness.
Innovation Solution
A method and apparatus utilize a first AI network with encoders and decoders to extract and enhance features from single or multiple data types, including cameras and LiDAR, using mapping networks to align and fuse these features into a unified bird's eye view representation, enabling the generation of high-definition maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If features from multiple data types (camera and LiDAR) are fused together, then the accuracy of high-definition map generation is improved, but the device complexity increases due to the need for multiple encoders and feature alignment mechanisms
Solution Approach 1:
The system divides the feature extraction process into separate encoders for different data types (camera encoder, LiDAR encoder). Each encoder independently processes its specific data type and extracts features, which are then aligned and fused. This segmentation allows the system to handle multiple data types without requiring a single complex encoder, thereby improving map generation accuracy while managing system complexity through modular design.
Solution Approach 2:
The patent introduces a feature alignment mechanism that acts as an intermediary between features extracted from different data types. This alignment module processes and coordinates the features before fusion, ensuring compatibility and consistency across different sensor modalities. The intermediary structure enables accurate feature integration without directly coupling the diverse input sources, thus improving accuracy while maintaining manageable system complexity.
2Adaptability or versatility
If a unified AI network is used to process both single and multi-type data, then the adaptability of the system is improved, but the difficulty of detecting and measuring increases due to the complexity of feature alignment
Solution Approach 1:
The patent designs a unified AI network architecture that can process both single-type and multi-type data inputs through a common framework. The network uses shared components (such as a common decoder and feature fusion mechanism) that operate regardless of the number or type of input data sources. This universal design enables the system to adapt to different sensing configurations (camera-only, LiDAR-only, or combined) without requiring separate processing pipelines, thereby improving adaptability while managing the complexity of feature alignment through standardized procedures.
Data Source
AI summary
A method of acquiring a high-definition (HD) map and an apparatus performing the method are disclosed. A method executed by an electronic device, according to one embodiment, may include acquiring first data including at least one type of data. The method may include acquiring a map image corresponding to the first data using a first artificial intelligence (AI) network based on the first data.