Black Ice Detection Using Convolutional Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for detecting black ice on road surfaces are costly, require dedicated infrastructure, and have limited accuracy, making them inefficient for widespread deployment and reliable detection, especially since black ice is a rare and hazardous condition that is hard to discern due to its transparent and matte appearance.
Innovation Solution
A cost-effective black ice detection system utilizing pre-trained convolutional neural networks to process images from existing road surface cameras, such as CCTV, to identify characteristic textures of black ice, allowing for automatic alerts without the need for dedicated lighting or new infrastructure, and can be activated on demand based on environmental conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If dedicated infrared imaging modules or temperature monitoring systems are deployed to detect black ice, then detection reliability is improved, but deployment cost and device complexity increase significantly
Solution Approach 1:
The patent applies universality by enabling existing road traffic cameras to perform dual functions: their original traffic monitoring role and the new black ice detection function. The convolutional neural network is trained to extract ice detection capabilities from standard camera footage, allowing one device to serve multiple purposes without requiring separate dedicated infrastructure.
Solution Approach 2:
The patent uses copying by creating a digital model (convolutional neural network) that learns to replicate the detection capabilities of expensive specialized sensors. Instead of deploying physical infrared or temperature sensors, the system copies the detection function through software that analyzes visual patterns in standard camera images, achieving similar reliability without specialized hardware.
2Area of stationary object
If permanent widespread dedicated black ice monitoring systems are deployed, then detection coverage is improved, but operational cost increases
Solution Approach 1:
The system leverages existing road traffic camera networks that are already permanently deployed across widespread areas for traffic monitoring. By adding black ice detection capabilities to these existing devices, the system achieves widespread coverage without the cost of installing new permanent infrastructure, eliminating the need for separate dedicated monitoring systems.
Solution Approach 2:
The system applies self-service by utilizing the existing infrastructure and power supply of road traffic cameras. These cameras are already operational, powered, and maintained for their primary function, so the black ice detection system piggybacks on this existing service infrastructure rather than requiring separate power sources and maintenance arrangements.
3Device complexity
If existing road traffic cameras are used for black ice detection, then deployment cost is reduced, but detection precision is limited
Solution Approach 1:
The patent applies parameter changes by transforming the way camera data is processed and interpreted. Instead of using standard image processing, the system changes the analytical parameters by applying convolutional neural networks with specific training parameters that focus on ice-related visual features. This transforms ordinary camera footage into precise detection data through advanced parameter-based analysis.
Solution Approach 2:
The patent substitutes mechanical/specialized sensor systems with an information processing system. Instead of relying on the physical properties of infrared or temperature sensors, the system uses computational algorithms (convolutional neural networks) to extract ice detection information from visual data, replacing hardware-based detection with software-based analysis that achieves comparable or superior precision.
4Measurement precision
If specialized equipment is used for black ice detection, then detection accuracy is improved, but ease of operation and maintenance deteriorates
Solution Approach 1:
The system maintains ease of operation by using familiar road traffic camera infrastructure that operators already know how to manage. The unified platform handles both traffic monitoring and ice detection through a single interface, eliminating the need for separate specialized equipment that would require separate operational procedures and maintenance protocols.
Data Source
Figure 1~2
Figure 3~5
Figure 6
AI summary
Embodiments include a black ice detection system, configured to identify the presence of black ice on road surfaces based on image processing of road surface images. The black ice detection system comprising: a memory; a processor coupled to the memory and the processor is configured to: retrieve a road surface image from one or more image capturing resources, wherein the road surface image includes an image of a physical road surface; respond to the retrieval of a road surface image by executing image processing on the retrieved road surface image as a real time road surface image. The image processing comprising: transforming the image of the physical road surface in the real time road surface image into a real time road surface texture value vector using a convolutional neural network, identifying a likelihood of black ice being present on the physical road surface included in the real time road surface image being reached by comparing the real time road surface texture value vector to both: an exemplar clear road surface texture value vector produced by the convolutional neural network based on one or more predetermined images of a road surface on which no black ice is formed; and an exemplar black ice road surface texture value vector produced by the convolutional neural network based on one or more predetermined images of a road surface on which black ice is formed, and when the comparison indicates a similarity threshold between the real-time road surface texture value vector and the exemplar black ice road surface texture value is satisfied, switching from a non-alert state to an alert state if, over a predetermined number of road surface images in the time series, the respective real time road surface value vectors exhibit increasing similarity to the exemplar black ice road surface value vector and decreasing similarity to the exemplar clear road surface value vector; the processor being further configured to: respond to the switch to alert state by transmitting a black ice alert to alert recipients of the black ice detection system.