Lane-Level Snow Depth Estimation Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems lack the ability to provide real-time, lane-level information on snow accumulation on roads, relying on macro-level reports that do not account for variations across different lanes and roads.
Innovation Solution
A vehicle system that captures images of the road, processes them using a scene segmentation neural network to create an edge map, and inputs this data along with a high-definition map into a trained regressor neural network to estimate snow depth for each lane, with aggregated data transmitted to edge and cloud servers for accurate and comprehensive snow depth mapping.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If macro-level snow accumulation reports are used, then information availability is improved, but measurement precision deteriorates because lane-level variations are not captured
Solution Approach 1:
The patent divides the road into multiple lanes and further segments each lane into sections, allowing snow depth measurement at lane-level granularity rather than macro-level geographic areas. This segmentation enables capture of spatial variations in snow accumulation across different lanes while maintaining information availability.
Solution Approach 2:
The patent transitions from macro-level geographic area measurements to lane-level measurements by adding spatial dimensionality. By using road geometry data and positioning information alongside image data, the system measures snow depth along the longitudinal and lateral dimensions of the road, enabling precise lane-level assessment.
2Measurement precision
If real-time lane level snow depth measurement is implemented, then measurement precision is improved, but device complexity worsens due to multiple sensors and processing systems
Solution Approach 1:
The patent makes the vehicle system multi-functional by using the same image sensor and processing system for both lane detection and snow depth measurement. The road geometry data serves multiple purposes: identifying lane boundaries, determining measurement locations, and correlating image pixels to physical snow depth. This universality reduces device complexity while maintaining lane-level precision.
Solution Approach 2:
The system uses the vehicle's own sensors, positioning data, and onboard computing resources to perform lane-level snow depth measurement without requiring external infrastructure. The vehicle self-services by processing its captured images through neural networks and correlating them with stored road geometry data, eliminating the need for dedicated snow measurement equipment.
3Reliability
If aggregated data from multiple vehicles is processed, then reliability is improved through averaging, but loss of time worsens due to data transmission and processing delays
Solution Approach 1:
The patent performs preliminary actions by pre-storing road geometry data and lane boundary information in the vehicle system before snow measurement is needed. During snow events, the system immediately correlates captured images with pre-loaded road geometry data, eliminating the need for real-time data retrieval and processing. This preliminary preparation reduces time loss while maintaining reliability through data aggregation.
Data Source
AI summary
A method comprises receiving an image of a road captured by a vehicle driving on the road, receiving a map of the road, the map comprising a road geometry of the road, obtaining an edge map of the road based on the image of the road, inputting the image, the map of the road, and the edge map into a trained regressor neural network, determining an estimated snow depth for each of one or more lanes of the road based on an output of the regressor neural network, and transmitting the estimated snow depth to an edge computing device.