Lane Boundary Prediction Using Non-Marking Road Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Real-world environment conditions such as rain, dirt, sand, fog, snow, etc., negatively affect the performance of advanced driver assistance system (ADAS) features by making lane detection more difficult due to the absence or obscurity of lane markings.
Innovation Solution
The use of machine learning models to predict lane boundaries by utilizing non-lane marking features such as lane widths, track edges, tire edges, taillight edges, etc., from image data, and confirming and updating these predictions using rearward-facing camera data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional lane marking detection methods are used, then the system is simple and easy to implement, but the detection reliability deteriorates in adverse environmental conditions such as rain, dirt, sand, fog, and snow
Solution Approach 1:
The patent segments the lane boundary detection problem into two distinct parts: visible lane boundaries detected through traditional image processing of lane markings, and predicted lane boundaries generated by machine learning models using alternative features. This segmentation allows the system to maintain simple traditional methods where they work while introducing complex AI methods only where needed, thus improving reliability without uniformly increasing complexity across all operating conditions.
Solution Approach 2:
The patent introduces machine learning models as an intermediary component that bridges the gap when traditional lane marking detection fails. These models use alternative features (vehicle tracks, tire marks, road geometry) as intermediate indicators to infer lane boundaries when direct lane marking observation is impossible due to environmental obstructions like rain, snow, or dirt.
2Reliability
If machine learning models are used to predict lane boundaries, then the detection reliability improves in adverse conditions, but the device complexity increases
Solution Approach 1:
The patent implements a dynamic detection system that adapts its methodology based on real-time environmental conditions and data quality. The system dynamically switches between traditional lane marking detection and machine learning-based prediction, and even dynamically updates the machine learning models using newly observed lane markings when visible. This dynamic approach ensures the system uses the most appropriate method for current conditions, improving reliability while managing complexity through adaptive behavior rather than static over-engineering.
Solution Approach 2:
The patent changes the feature parameters used for detection based on availability. When lane markings are visible, the system uses lane marking color, position, and continuity parameters. When lane markings are obscured by environmental conditions, the system transitions to using alternative parameters such as vehicle track patterns, tire mark locations, road edge geometry, and contextual spatial relationships. This parameter substitution strategy maintains detection reliability across varying conditions without requiring a completely different system architecture.
3Measurement precision
If only visible lane markings are used for detection, then the system complexity is low, but the measurement precision deteriorates when lane markings are obscured or absent
Solution Approach 1:
The patent performs preliminary actions by continuously training and updating the machine learning models with observed lane marking data even when markings are visible. This preliminary training ensures the models are ready and calibrated to provide accurate predictions when lane markings become obscured, maintaining measurement precision without waiting for failure conditions. The system proactively builds predictive capability in advance, reducing the complexity of emergency responses when detection becomes difficult.
Data Source
AI summary
A system for determining lane information. A memory storing instructions executable by a processor includes instructions to receive forward image data of a roadway from a forward-facing camera of a vehicle, determine visible lane boundaries of the driving lane based on lane marking features in the forward image data, determine a predicted lane boundary of the driving lane based, at least in part, on forward features in the forward image data, and determine a driving path of the vehicle through the visible lane boundaries and the predicted lane boundary.