Road Curvature Prediction Using Machine Learning and Trace Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for calculating road curvature, particularly at intersections and complex road junctions, face challenges such as missing data, errors in map geometry, and difficulties in fitting splines, leading to inaccurate curvature calculations.
Innovation Solution
A machine learning approach is employed to predict road curvature by collecting location trace data from vehicles, dividing the road network into bounded areas, extracting features like vehicle density and heading, and training a supervised machine learning model using ground truth curvature values to provide independent curvature data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional spline-based methods are used to calculate road curvature from map geometry, then curvature data can be obtained for simple road segments, but the method fails at intersections and complex road junctions where curvature data is missing or inaccurate
Solution Approach 1:
The patent replaces the traditional mechanical/mathematical spline-fitting approach with a machine learning-based system. Instead of relying on geometric constructions that fail at intersections, the system uses trained models that process location trace data to predict curvature values even at complex road junctions where traditional methods cannot operate.
Solution Approach 2:
The patent introduces machine learning models as an intermediary between location trace data and curvature values. The models are trained on areas with ground truth curvature data and then applied to predict curvature in areas where traditional calculation methods fail, particularly at intersections and complex road junctions.
2Ease of manufacture
If map geometry data is used as the basis for curvature calculation, then curvature can be derived from existing map attributes, but errors in map geometry lead to inaccurate curvature calculations
Solution Approach 1:
The system uses location trace data from vehicles themselves to calculate curvature, rather than relying on externally provided map geometry. By processing actual vehicle movement data, the system generates curvature information that is independent of potential errors in map geometry, allowing vehicles to self-determine road characteristics.
Solution Approach 2:
The patent substitutes the map-geometry-based calculation mechanism with a data-driven machine learning approach. Instead of deriving curvature from potentially erroneous map attributes, the system uses trained models that process location trace data, replacing the vulnerable geometric calculation chain with a more robust statistical approach.
3Measurement precision
If location trace data from vehicles is collected and processed, then accurate curvature prediction is achieved, but significant computational resources and data processing are required
Solution Approach 1:
The patent performs the computationally intensive work of training machine learning models in advance, using location trace data and ground truth curvature values. Once trained, the models can predict curvature with high accuracy using minimal real-time computation, thereby resolving the contradiction between accuracy and computational complexity during actual operation.
Data Source
AI summary
An approach is provided for generating road curvature data. The approach, for example, involves map matching location trace data to a road network, wherein the location trace data is associated with ground truth curvature values. The approach also involves dividing the road network into a plurality of bounded areas (e.g., boxes or other shapes). The approach further involves extracting one or more training features for each bounded area of the plurality of bounded areas from the location trace data map matched to said each bounded area. The approach further involves training a machine learning model based on the one or more training features and ground truth curvature values. The approach further involves providing the trained machine learning model to predict the road curvature data.