Online Vectorized Mapping for Autonomous Vehicle BEV Lane Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Creating affordable and precise vectorized maps for autonomous vehicles, particularly in urban environments, is challenging due to the computational intensity of extracting polylines from polygon contours, and existing methods are inefficient and wasteful.
Innovation Solution
A system using an image sensor and an Electronic Control Unit (ECU) with a semantic segmentation deep learning model and vectorization post-processing module to convert image frames into a Bird's Eye View (BEV) map, extracting polygon contours and center polylines to create vectorized maps, which can be uploaded to a server for access by other vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional HD maps are used for autonomous driving, then mapping precision is improved, but manufacturing cost and update frequency are worsened
Solution Approach 1:
The system enables autonomous vehicles to self-generate vectorized maps by detecting and extracting map features directly from sensor data during normal operation. Each vehicle acts as its own mapping device, eliminating the need for expensive professional mapping teams while maintaining continuous map updates through real-time feature extraction from camera and sensor inputs.
Solution Approach 2:
Instead of creating expensive physical HD maps through specialized mapping equipment, the system creates digital copies of map features by detecting and vectorizing visual elements from standard vehicle sensors. The vectorized map serves as a simplified digital representation that captures essential navigation information without requiring costly original mapping operations.
2Productivity
If vectorized maps are created in real-time during autonomous driving, then map update frequency is improved, but computational complexity is worsened
Solution Approach 1:
The system extracts only the essential polyline features needed for navigation from the full sensor data stream, rather than processing and storing complete HD map data. By focusing extraction on critical linear features like lane markings and curbs, the computational burden is significantly reduced while maintaining map utility for autonomous driving decisions.
Solution Approach 2:
The system transforms complex sensor data into simplified vectorized representations by changing the data parameters from raw pixel information to geometric polyline descriptions. This parameter transformation reduces data dimensionality and computational requirements while preserving the essential spatial information needed for navigation and map updates.
3Manufacturing precision
If polylines are extracted from polygon contours for vectorized mapping, then vectorized map quality is improved, but processing time and computational effort are worsened
Solution Approach 1:
The system performs preliminary detection and segmentation of map features during normal sensor processing before the actual polyline extraction occurs. By pre-identifying potential map features and organizing them into candidate polygons, the subsequent polyline extraction step operates on already-prepared data structures, significantly reducing the time required for high-precision vectorized map generation.
Data Source
AI summary
A system for creating online vectorized maps for autonomous vehicles includes an image sensor and an electronic control unit (ECU). The image sensor captures a series of image frames. The ECU includes a memory, a central processing unit (CPU), and a transceiver. The memory stores a semantic segmentation deep learning model and a vectorization post-processing module as computer readable code. The CPU executes the semantic segmentation deep learning model and the vectorization post-processing module to output a vectorized map of an external environment of a vehicle. The transceiver uploads the vectorized map to a server such that the vectorized map can be accessed by a second vehicle that uses the vectorized map to traverse the external environment.