Road Model Estimation Using Grid-Map Boundary Scanning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for recognizing road boundaries in vehicle environments are prone to errors, particularly in urban settings, and are limited to simple surroundings like highways, failing to accurately identify the actual roadway boundaries and handle intersections or complex scenarios.
Innovation Solution
A method involving the creation of a road model through sensor data fusion, using enlarged grid cells to ensure sufficient space for vehicle passage, and orthogonal scanning to accurately determine road boundaries, incorporating various sensor types and weighted path analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex algorithms are used to identify prominent points and classify them as road boundary points, then road boundary detection capability is improved, but device complexity and computational requirements increase
Solution Approach 1:
The patent segments the road boundary detection task into distinct phases: first identifying prominent points through clustering, then classifying them as left or right boundary points, and finally fitting a mathematical model. This segmentation transforms a single complex classification problem into multiple simpler sequential steps, reducing overall algorithmic complexity while maintaining detection accuracy.
Solution Approach 2:
The patent introduces intermediate structures including clusters of prominent points and mathematical models (Euler spiral, polynomial) as mediators between raw sensor data and final road boundary representation. These intermediaries simplify the classification task by providing structured representations that capture essential road geometry features without requiring direct complex classification of all raw points.
2Manufacturing precision
If mathematical models like Euler spiral are used to describe road edges, then road profile representation is improved, but adaptability to diverse road scenarios deteriorates
Solution Approach 1:
The patent employs dynamic model selection where the mathematical model type (Euler spiral, polynomial, or other curves) is chosen based on the specific road scenario and detected prominent points. Rather than using a fixed model, the system adapts the model type and parameters to match the actual road geometry, enabling accurate representation of highways, urban streets, intersections, and junctions.
Solution Approach 2:
The patent changes model parameters dynamically based on the road scenario. Different mathematical models with varying degrees of freedom and geometric properties are selected and configured according to the detected prominent points and road characteristics. This parameter adaptation allows the same framework to accurately model diverse road types from straight highways to complex urban intersections.
3Measurement precision
If prominent points are extracted from sensor data, then road boundary identification is improved, but reliability deteriorates due to false identification of highly reflective objects
Solution Approach 1:
The patent segments prominent points into clusters based on spatial proximity and characteristics, then processes each cluster separately to identify boundary points. This clustering approach distinguishes between isolated highly reflective objects and continuous road boundary structures, improving reliability by requiring spatial consistency rather than relying on single prominent points.
Solution Approach 2:
The patent implements feedback through iterative refinement where initial prominent point extraction is followed by clustering analysis, then classification, and finally model fitting that can identify and correct erroneous classifications. The system uses the fitted road model as feedback to validate whether detected boundary points are consistent with the overall road geometry, correcting misidentified highly reflective objects.
Data Source
AI summary
A method for creating a road model for planning a trajectory of an ego vehicle, includes capturing a vehicle environment with at least one environment capturing sensor. Objects are detected in the vehicle environment. The method includes producing a grid map with a plurality of grid cells and entering the detections into the grid map wherein the detections are enlarged prior to being entered. The method also includes determining at least one path through the grid map which consists only of vacant grid cells. The at least one path is gradually scanned, wherein detections are determined orthogonally to a scanning direction of the path on both sides. The method further includes assigning the detections along the at least one path to road boundaries and creating the road model based on the previously determined information.