Lane-Level Road Map Generation with Curve-Fitted Sensor Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating precise lane map data in autonomous vehicles face challenges such as indistinct lane markings, weather obstructions, and the need for highly accurate pre-mapped terrain features, which existing technologies struggle to address effectively, especially in dynamic environments like construction sites and seasonal changes.
Innovation Solution
An in-vehicle system combining a GPS receiver, inertial sensor, and camera to acquire and improve positional data through curve fitting, allowing for precise lane-level road map generation, with the ability to update maps with transient road features by collecting data from multiple vehicles and transmitting it to a central repository for post-processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If computer vision is used to detect road or lane boundaries, then lane detection can be performed, but detection reliability deteriorates when lane markings are indistinct, obscured by weather or obstructions, or when lighting conditions are unfavorable
Solution Approach 1:
The patent introduces an intermediary system that combines multiple sensors (GPS, inertial sensors, cameras) to detect and track lane boundaries. When visual detection fails due to poor lighting, weather, or obscured markings, the system uses GPS and inertial data as intermediary measurements to maintain lane detection reliability, effectively mediating between direct visual detection and fallback methods.
Solution Approach 2:
The system dynamically changes detection parameters based on environmental conditions. When lighting conditions deteriorate or lane markings become indistinct, the system switches from relying primarily on visual parameters to using GPS coordinate parameters and inertial measurement parameters, thereby maintaining detection reliability across varying conditions.
2Measurement precision
If precision GPS (DGPS or equivalent) augmented by inertial sensing is used to keep the vehicle in the lane, then positioning accuracy can be improved, but the requirement for highly accurate pre-mapped terrain features increases system complexity
Solution Approach 1:
The patent merges GPS reception, inertial sensing, and visual detection into a unified lane positioning system. By combining these multiple measurement sources and processing them together through a single processor, the system achieves high positioning accuracy without requiring separate complex subsystems, thereby reducing overall system complexity while maintaining precision.
Solution Approach 2:
The system uses a multi-functional processor that handles GPS data processing, inertial sensor fusion, visual image processing, and lane boundary detection all through a single device. This universal approach eliminates the need for multiple specialized hardware systems, reducing device complexity while maintaining measurement precision across all functions.
3Measurement precision
If 3D simultaneous localization and mapping using 3D range sensing is used, then vehicle position relative to surrounding terrain features can be determined, but the requirement to map 100% of road surround to 2 cm to 10 cm accuracy increases measurement precision requirements
Solution Approach 1:
The patent extracts only the essential lane-level positioning information needed for autonomous driving rather than mapping the entire surrounding environment. By taking out and focusing specifically on lane boundary detection and vehicle position relative to lanes, the system achieves required precision without the complexity of comprehensive 360-degree environmental mapping.
Solution Approach 2:
Instead of mapping 100% of the road surround to high accuracy, the system performs partial mapping focused specifically on lane boundaries and relevant road features. This partial action approach provides sufficient positioning precision for lane-keeping while dramatically reducing the complexity and data requirements of full environmental mapping.
4Area of stationary object
If aerial and satellite based road mapping is used, then mapping coverage can be achieved, but mapping accuracy deteriorates because it cannot reach 2 cm to 10 cm precision required for lane-level positioning
Solution Approach 1:
The patent transitions from two-dimensional aerial/satellite imagery to a multi-dimensional sensor fusion approach combining GPS coordinates, three-axis inertial measurements, and multi-camera visual data. By adding these additional measurement dimensions, the system achieves lane-level precision (2 cm to 10 cm accuracy) that cannot be obtained from satellite-based two-dimensional mapping alone.
Data Source
AI summary
An in-vehicle system for generating precise, lane-level road map data includes a GPS receiver operative to acquire positional information associated with a track along a road path. An inertial sensor provides time local measurement of acceleration and turn rate along the track, and a camera acquires image data of the road path along the track. A processor is operative to receive the local measurement from the inertial sensor and image data from the camera over time in conjunction with multiple tracks along the road path, and improve the accuracy of the GPS receiver through curve fitting. One or all of the GPS receiver, inertial sensor and camera are disposed in a smartphone. The road map data may be uploaded to a central data repository for post processing when the vehicle passes through a WiFi cloud to generate the precise road map data, which may include data collected from multiple drivers.