Development of a Human Activity Recognition Algorithm Based on BiLSTM for Construction Workers

The safety management of a water conservancy construction site faces difficulties. To effectively supervise all workers' operational activity, this study develops a human activity recognition model using a modified BiLSTM algorithm with attention mechanism. The model has an accuracy of 91.65%, and a precision of 91.76%. The model was deployed to a cloud server and a mobile app was developed to transmit the collected action data. A validation test was performed in construction site, and the results show that the proposed model can accurately recognize the workers' activity.