A System to Detect Human Fall-off from an Altitude

Abstract

Working at heights continues to be one of the leading causes of death and serious injuries. Falls from an altitude are the second most prevalent cause of injury-related fatality following traffic accidents. After a fall accident, if it is not reported in time and the victim does not receive first aid, this delay can lead to death, especially when there is no onlooker to report it. This study proposes a smartphone-based practical and cost-effective solution for detecting human fall-offs from an altitude. The proposed approach uses built-in sensors available in modern smartphones, such as an accelerometer and barometer. These sensors provide helpful data that can be utilized for accurately predicting the likelihood of falls. A Support Vector Machine (SVM)-based system has been trained and tested using the "change in altitude" and "absolute linear acceleration" to detect the fall-off event. The SVM-based model is highly accurate with accuracy, precision, recall, F1-score and AUC of 94.33%, 0.928, 0.935, 0.937, and 0.952, respectively.  The research results suggest that this system can be used to determine fall prevention and mitigation techniques in various safety programs worldwide.