Action Recognition Based on 3D Skeleton and LSTM for the Monitoring of Construction Workers’ Safety Harness Usage

Abstract

Fall from height (FFH) is the most common construction accident in the construction industry, thus it is significant to monitor the use of safety harnesses, which are critical to the prevention of FFH. Sensing or computer vision technologies have been adopted to identify workers’ safety harness usage. However, previous research focused mainly on whether a worker wears a safety harness rather than on whether he or she properly fixes it to a lifeline, which is vital to prevent FFH but difficult to monitor. This research establishes an action recognition method based on a three-dimensional (3D) skeleton and long short-term memory (LSTM) to aid in automatically monitoring whether safety harnesses are fixed properly on site. An indoor experiment, which considered the features of a common real construction scenario—working on scaffolding—was conducted to test the effectiveness and feasibility of the proposed method. The result shows that the method achieves an acceptable precision and recall rate and can be used to detect the incorrect use of safety harnesses by combining multiple actions. This will contribute to the prevention of FFH in practice as well as to the body of knowledge of construction safety management.