Artificial Intelligence-Driven All-Terrain Vehicle Crash Prediction and Prevention System

Abstract

HighlightsAn AI-driven system for predicting and preventing ATV crashes was developed.Machine learning model achieved rollover prediction accuracy of over 99%.The system has the potential to significantly reduce ATV-related injuries and fatalities by enabling preemptive actions.Abstract.All-Terrain Vehicle (ATV) crashes have become a public health concern in the U.S. over the past decades, resulting in numerous fatalities and hospitalizations. Most of those incidents could have been prevented if riders could better assess their ability to handle risks. Currently, risk factors associated with ATV incidents have already been studied. However, little effort has been made toward developing practical applications that assist the rider in preventing crashes. Commercial ATV safety systems, such as Farm Angel, focus on post-crash detection and emergency medical services (EMS) alerting rather than preventive measures. Machine learning prediction models can be used to assist riders in taking preventive measures to avoid an imminent crash. In this study, we developed a system that leverages the predictive power of machine learning algorithms to assess the likelihood of a crash in real-time and alert the riders, thus allowing them to prevent the crash. To the best of our knowledge, this is the only system ever developed for ATVs specifically that can predict rollover incidents. The crash likelihood is estimated by a deep neural network that considers the ride parameters (e.g., ATV speed, turning radius, and roll and pitch angles), ATV characteristics (e.g., width, length, wheelbase), and human factors (i.e., presence of a rider). The ATV characteristics and the presence of a rider are retrieved from the rider’s input through a smartphone application developed specifically for this study. The ride parameters are retrieved from an embedded system (attached to the ATV). Validation and performance tests indicated that: (1) the proposed device has a rollover prediction system with an accuracy superior to 99%; (2) the system can detect roll and pitch angles with average errors of 0.26 and 0.54 degrees, respectively; and (3) the system can detect the ATV’s speed with an average error of 0.75 m s-1. Keywords: ATV safety, Crash prediction, Machine learning, Rollover prevention.