Road Defect Prediction Using Vision Transformer Depth Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for road defect level prediction are time-consuming and costly, requiring manual measurement of each road, which is inefficient for large-scale road infrastructure maintenance and does not account for the rapid degradation of roads over time.
Innovation Solution
A computer-implemented method using a vision transformer model to obtain depth maps and a semantic segmentation model to detect road pixels, which are then used to predict road defect levels by fitting the data into a road surface model, enabling efficient and scalable prediction of road defects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual measurement methods are used to assess road defects, then measurement precision is improved, but productivity deteriorates
Solution Approach 1:
The patent replaces manual mechanical measurement systems with an automated computer vision system comprising image capture devices, processing units, and algorithms. The system automatically captures road images, processes them through multiple analysis modules (depth estimation, semantic segmentation, defect detection), and generates defect level predictions without human intervention, thereby maintaining measurement precision while dramatically improving productivity
Solution Approach 2:
The patent creates digital copies of road surfaces through high-resolution imaging and depth mapping. By capturing visual representations and processing them through computational models, the system generates virtual replicas of road conditions that can be analyzed without physical contact or manual measurement, enabling rapid assessment while preserving measurement accuracy
2Loss of information
If manual road inspection is conducted to cover every road, then measurement completeness is improved, but loss of time worsens
Solution Approach 1:
The patent implements a dynamic inspection system that can adapt its assessment scope and depth based on priorities, resources, and road criticality. The system enables selective intensive inspection of high-priority roads while performing rapid screening on others, allowing comprehensive coverage across the entire road network within reduced timeframes through flexible, adaptive resource allocation
Solution Approach 2:
The patent divides the road network into segments that can be independently assessed and prioritized. By segmenting the inspection task into manageable units (individual road sections, specific defect types, priority levels), the system can process multiple segments simultaneously or in optimized sequences, ensuring complete coverage while minimizing total inspection time through parallel processing
3Reliability
If frequent road inspections are performed to monitor rapid degradation, then reliability is improved, but productivity deteriorates
Solution Approach 1:
The patent implements periodic automated inspections at scheduled intervals, allowing rapid re-assessment of road conditions without the logistical overhead of manual inspection teams. The system can be deployed repeatedly and efficiently, enabling frequent monitoring cycles that maintain high reliability for detecting rapid road degradation while preserving productivity through automation
Solution Approach 2:
The patent creates a self-sufficient inspection system that autonomously captures images, processes data, detects defects, and generates reports without requiring human inspectors for each assessment. The automated system performs all inspection functions independently, enabling frequent monitoring at low marginal cost and high efficiency, thereby resolving the contradiction between reliability and productivity
Data Source
AI summary
Systems and methods for road defect level prediction. A depth map is obtained from an image dataset received from input peripherals by employing a vision transformer model. A plurality of semantic maps is obtained from the image dataset by employing a semantic segmentation model to give pixel-wise segmentation results of road scenes to detect road pixels. Regions of interest (ROI) are detected by utilizing the road pixels. Road defect levels are predicted by fitting the ROI and the depth map into a road surface model to generate road points classified into road defect levels. The predicted road defect levels are visualized on a road map.