The construction industry experiences the highest rate of casualties from safety-related accidents at construction sites despite continuous social interest in safety management. Accordingly, various studies have been conducted on safety management, wherein recent studies have focused on its integration with Machine Learning (ML). In this study, we proposed a technology for recognizing struck-by hazards between construction equipment and workers, where a Convolutional Neural Network (CNN) and sound recognition were combined to analyze the changes in the Doppler effect caused by the movements of a subject. An experiment was conducted to evaluate the recognition performance in indoor and outdoor environments with respect to movement state, direction, speed, and near-miss situations. The proposed technology was able to classify the movement direction and speed with 84.4-97.4% accuracy and near-misses with 78.9% accuracy. This technology can be implemented using data obtained through the microphone of a smartphone, thus it is highly applicable and is also effective at ensuring that a worker becomes aware of a struck-by hazard near construction equipment. The findings of this study are expected to be applicable for the prevention of struck-by accidents occurring in various forms at construction sites in the vicinity of construction equipment.
Keywords: Convolutional Neural Network; Doppler effect; construction safety; high-frequency sound; struck-by accident.