An Accelerometer-Based Two-Wheeler Fall Detection Analysis

Vehicle Activity Recognition (VAR) involves identifying changes in a vehicle's behavior within a fixed time frame under specific conditions. Powered Two-Wheeler (PTW) fall detection is a critical VAR problem, where the goal is to monitor the vehicle and detect fall events using various sensors and cameras. However, the high risks associated with collecting real-world PTW fall data has resulted in limited amount of available data for building accurate recognition models. In this work, we leverage the publicly accessible datasets created especially for PTW fall detection. This dataset contain accelerometer and gyrometer data recorded over a certain time interval. We have labeled the data into four classes. Many deep learning and machine learning techniques have been used in our analysis. Of the evaluated techniques, Random Forest exhibited the highest efficacy in accurately identifying fall incidents, highlighting its potential for effective PTW fall detection.