Wearable-based Activity Recognition of Construction Workers using LSTM Neural Networks

Identification of worker behaviors may be used to quantify and monitor performance in an intelligent construction system employing employees and delivering onsite training through augmented reality. This research aims to present a technique for recognizing construction worker movement utilizing Inertial Measurement Unit (IMU) sensors from wearable devices. The raw IMU data are put into a deep learning model termed a Long Short-Term Memory (LSTM) neural network for automated feature extraction, producing a time-dependent high-level feature vector. The vector is then utilized to identify worker activities. To assess the proposed deep learning model, sensor data from worker construction projects were gathered in the VTT-ConIot dataset. These sensor data were collected using a tri-axial accelerometer, a tri-axial gyroscope, and a triaxial magnetometer placed in a wearable device carried by construction workers at several body positions (hip, back, and hand). The performance of the model is quantified using a variety of measures, including accuracy, precision, recall, and F1-score. According to experimental findings, the suggested LSTM model attained the best accuracy of 97.32% when sensor data from construction workers' back positions were used. Additionally, the findings suggest that several sensors may be employed to boost identification performance.