Sound-Based Construction Activity Monitoring with Deep Learning

Automated construction monitoring assists site managers in managing safety, schedule, and productivity effectively. Existing research focuses on identifying sounds to determine the type of activity. However, there are two major limitations: inability handle a mixed sound environment which multiple activity occur simultaneously, precisely locate start end times each individual This aims fill this gap through developing an innovative deep learning-based method. The proposed model combines benefits Convolutional Neural Network (CNN) for extracting features Recurrent (RNN) leveraging contextual information environments with polyphony noise. In addition, dual threshold output permits exact identification finish timings activities. Before training testing collected from modular factory, has been pre-trained publicly available general event data. All designs have confirmed by ablation study, extended experiments were also performed verify versatility present additional or great potential be used autonomous