Road Surface Abnormality Detection Using Predictive Model Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing road surface abnormality detection technologies can only estimate abnormalities defined by labels during the learning process, failing to detect unforeseen abnormalities without prior learning.
Innovation Solution
A road surface abnormality detection apparatus that acquires chronological observation data, uses a trained model to predict road surface conditions, and detects abnormalities by comparing observed data with predicted data, eliminating the need for predefined labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a trained model is used to predict road surface data, then the detection of predefined abnormalities is improved, but the ability to detect unforeseen abnormalities deteriorates
Solution Approach 1:
Instead of using a trained model to classify abnormalities directly (which only detects predefined types), the patent inverts the approach by using an autoencoder to reconstruct normal road surface data. The detection mechanism then identifies abnormalities by comparing actual data with the reconstructed normal data, allowing detection of any unforeseen abnormalities without requiring prior learning of specific abnormal patterns.
Solution Approach 2:
The patent extracts the normal behavior pattern from the data using an autoencoder, separating the concept of 'normal' from 'abnormal'. By taking out and storing only the normal data patterns in the autoencoder, the system can detect any deviation from normality without needing to be trained on specific abnormal conditions, thus enabling detection of unforeseen abnormalities.
2Measurement precision
If learning is performed for specific abnormality types, then the detection of those specific abnormalities is improved, but the complexity of the training model increases
Solution Approach 1:
The patent extracts and stores only normal road surface data patterns in the autoencoder during training. By taking out and encoding only the normal data, the model remains relatively simple without needing to learn multiple specific abnormal patterns. The complexity is reduced because the model only needs to capture normal behavior, not distinguish between different types of abnormalities.
Solution Approach 2:
The autoencoder performs self-supervised learning by attempting to reconstruct the input normal data. This self-service mechanism allows the model to learn normal patterns automatically without requiring labeled abnormal data or complex training procedures. The system serves itself by using the normal data to teach it what normality looks like, eliminating the need for complex anomaly-specific training.
Data Source
AI summary
A road surface abnormality detection apparatus includes a data acquisition unit, a prediction unit, and an abnormality detection unit. The data acquisition unit acquires road surface observation data obtained by observing a road surface in a chronological order. The prediction unit predicts road surface data at a first timing from a plurality of road surface observation data at a plurality of respective timings earlier than the first timing by using a trained model generated in advance through machine learning. The abnormality detection unit detects, when a difference between road surface observation data at the first timing and the predicted road surface data at the first timing is equal to or larger than a predetermined threshold, an abnormality of the road surface.