A Road Condition Classifier via Lock Embedded IMU on Dock-Less Shared Bikes

Abstract

Shared bikes have been gaining popularity worldwide in recent years, and the number of riders has also been increasing rapidly. However, not all riders have the same riding abilities, and these differences can potentially pose a risk to beginners. In addition, while certain types of bikes such as mountain bikes can handle bumps and holes, the uncertain road conditions might make it uncomfortable and even dangerous for riders of shared bikes. To solve this problem, road conditions should be detected by some effective methods and uploaded to online maps so that riders can choose routes which are suitable for their riding preferences. Therefore, we designed a road classifier based on a lock-embedded inertial measurement unit (IMU) on shared bikes with enabled road surface detection while riding. For training and evaluating the system, 20 subjects were recruited to collect data on dock-less shared bikes with an embedded IMU. To accurately classify road conditions, first, data rotation and feature extraction were performed. Then, linear discriminant analysis (LDA) was used to establish a final model. Cross-validation was performed and showed the accuracy of the model of classifying asphalt road, pebble path, and bumpy path pavement was 95.3%, which showed promising potential in the information expansion of an online mapping which could significantly enhance rider experience.