Road Surface Monitoring System Through Machine Learning Ensemble Models

The proliferation of large urban centers presents numerous challenges, including effective road pavement maintenance, which is a critical issue. To address this concern, we developed a low-resource framework that utilizes Inertial Measurement Units (IMUs) and elementary Machine Learning (ML) classification models to monitor surface defects in the pavement. Our platform recorded accelerometer and gyroscope measurements on a test vehicle's damped and undamped mass while driving on Mexico City's street. We then labeled these measurements to identify and classify general and specific elements of road irregularities, including smooth and uneven road segments, manholes-potholes, speed bumps, and patches. We employed a time series analysis and feature extraction to preprocess the data in both the time and frequency domains. Next, utilizing an exhaustive grid search methodology, we trained elementary classification algorithms to find the best possible predictors and train the ensemble models with them. Based on several validation routes traveled, the overall system demonstrated a macro F1-score of 0.827. Finally, we deployed the algorithms and models onto a cloud instance to process incoming raw data, with resultant predictions stored in a database that can be visualized on a web platform.