Road Damage Diagnosis Using Acceleration Data Collected from On-Board Devices Placed at Different Locations Over Multiple Periods

Abstract

In recent years, research has been conducted to develop a simplified method for inspecting road pavement using probe data, as the coverage of road patrolling and road surface condition surveys remains limited. The authors have proposed a method to detect road pavement damage using LSTM, a deep learning model well-suited for analyzing time-series data, applied to probe data from a single time period. However, a major drawback of using single-period probe data is that it may misidentify damage when the vehicle does not pass directly over the damaged area. On the other hand, with multi-temporal probe data, variations in acceleration trends due to differences in sensor placement make it challenging to accurately assess road damage.

This study proposes a method for assessing road pavement damage using multi-temporal probe data, incorporating acceleration correction. As a result, road pavement damage could be determined with an accuracy of F value of 0.8 or higher.