Road Hazard Detection Training Data Using Telematics and Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing road analysis methods fail to accurately and efficiently detect road hazards like potholes, leading to false-positive or false-negative detections, which can delay repairs and increase accident risks.
Innovation Solution
A method for providing training data for neural networks that integrates telematic data with image data to identify road hazards, weighting more relevant hazards, and automating the selection of training data to enhance accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If automated street monitoring systems are implemented to continuously monitor road conditions, then detection time is improved and areas in need of repair can be identified, but the system cannot achieve sufficiently accurate and complete analyses of road hazards
Solution Approach 1:
The patent combines image data from cameras with telematic data from vehicle sensors to create a multi-source detection system. This merging of different data types enables both timely detection through automated monitoring and accurate identification of road hazards by cross-validating information from multiple sources
Solution Approach 2:
The patent introduces a neural network as an intermediary that processes and analyzes the combined image and telematic data. This intermediary component bridges the gap between raw data collection and accurate hazard identification, enabling the system to achieve both speed and precision
2Adaptability or versatility
If conventional training data based on image data alone is used, then all road hazards in the image data are given equal weight, but this leads to suboptimal training where less relevant road hazards are over-represented
Solution Approach 1:
The patent applies local quality by differentiating the weight and importance of different road hazards based on their relevance to vehicle safety. Rather than treating all hazards equally, the system assigns different weights to hazards based on collision probability and severity, making the training data more targeted and effective
Solution Approach 2:
The patent changes the parameter of hazard weighting by integrating telematic data to determine the relevance and importance of different road hazards. This parameter change transforms the training approach from uniform weighting to differentiated weighting based on actual risk assessment
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A computer-implemented method for providing training data for training a neural network for road analysis comprises: obtaining initial telematic data associated with a drive of a vehicle on a surface; obtaining initial image data of the surface, wherein obtaining the initial image data is associated with the drive of the vehicle; and selecting at least a portion of the initial image data as the training data, based at least partly on the initial telematic data.