Deep learning-based slip-trip falls and near-falls prediction model using a single inertial measurement unit sensor for construction workplace

Slip-trip falls are one of the leading causes of injury in dynamic constructive workplace environments. To solve the problem of slips-trip in the construction industry, it is important not only to predict falls in order to be able to apply various fall prevention systems, but also to detect and monitor slip-trip near-falls in order to eliminate various dangerous anomalies in a timely manner. In this study, we propose a deep learning-based method for predicting and classifying activities of daily living (ADLs), trip falls, slip falls, trip near-falls and slip near-falls using a waist-attached wearable device with an embedded single inertial measurement unit (IMU) sensor containing a 3-axis accelerometer and a gyroscope. A total of 34 young and healthy participants took part in experiment to collect accelerometer and gyroscope data while performing 8 types of ADLs, slip-trip falls, and slip-trip near-falls. The data was processed and then 30 features were extracted from them. The resulting feature data was then transformed using bicubic interpolation to fit the input data size of the modified interpretation of LeNet-5 Convolutional Neural Network (CNN) deep learning algorithm. After completing 5-folds cross validation with 5 evaluation criteria, the proposed prediction and classification method showed high performance with an accuracy of about 0.9039, a specificity of about 0.9753 and F1-score of about 0.9049. Therefore, it is believed that the method proposed in this study can be used not only to increase accuracy by reducing false alarms of systems for early detection and prevention of slip-trip falls, but also for various systems for detecting and timely elimination of dangerous anomalies in the construction workplace.