Road Friction Estimation via Camera Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for estimating road friction rely on one-dimensional sensors and physics-based models, resulting in low availability and unacceptably low confidence, especially when real-time friction estimates are sporadic and predictive ability is hampered, compromising driver safety.
Innovation Solution
The method involves pre-processing forward-looking camera images, segmenting them into patches, transforming into a bird's eye view, quantizing, and classifying to generate a road friction estimate, which can be used to provide driver information, control vehicle motion, and inform cloud-based alert services, enhancing predictive ability and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If one-dimensional on-vehicle sensors and physics-based models are used to estimate road friction, then the system complexity is low, but the availability and confidence of friction estimates are unacceptably low
Solution Approach 1:
The patent replaces physics-based mechanical models with computer vision and machine learning approaches. Specifically, it uses forward-looking camera images processed through convolutional neural networks to estimate road friction coefficients, substituting optical sensing and data-driven algorithms for traditional mechanical sensors and physics equations.
Solution Approach 2:
The patent changes the measurement parameters from direct mechanical sensing to image-based feature extraction. It transforms road surface visual characteristics (texture, color, patterns) captured by cameras into friction coefficient estimates through learned mappings, rather than measuring physical forces or accelerations.
2Measurement precision
If forward-looking camera images are processed through pre-processing, patch segmentation, BEV transformation, quantization, and classification, then the accuracy and predictive ability of friction estimates improve, but the computational complexity increases
Solution Approach 1:
The patent divides the forward-looking camera image into multiple patches or regions of interest before processing. This segmentation allows the system to focus computational resources on specific areas that contain relevant friction information, improving accuracy while managing computational load through localized analysis.
Solution Approach 2:
The patent transforms the image from a forward-looking perspective to a bird's-eye view (BEV) representation. This dimensional transformation reorganizes the spatial information to better align with how friction varies across the road surface, improving measurement precision by presenting data in a more analytically useful format.
3Reliability
If physics-based models are used for real-time friction estimation, then the system is simple to implement, but the predictive ability is hampered and estimates are sporadic
Solution Approach 1:
The patent performs preliminary processing of camera images including pre-processing, patch segmentation, and BEV transformation before friction estimation. These preparatory steps organize and enhance the input data, enabling more accurate and consistent friction predictions while maintaining real-time capability through efficient pipeline design.
Solution Approach 2:
The patent creates a virtual representation of the road surface friction conditions by processing and transforming camera image data. This digital copy of the physical road state provides continuous, predictable estimates without requiring direct physical measurement, improving reliability while keeping the system implementable through software processing.
Data Source
AI summary
Methods and systems for generating and utilizing a road friction estimate (RFE) indicating the expected friction level between a road surface and the tires of a vehicle based on forward looking camera image signal processing. A forward-looking camera image is pre-processed, patch segmented (both laterally and longitudinally, as defined by wheel tracks or the like), transformed into a bird's eye view (BEV) image using perspective transformation, patch quantized, and finally classified. The resulting RFE may be used to provide driver information, automatically control the associated vehicle's motion, and/or inform a cloud-based alert service to enhance driver safety.


