Smart Construction: A Deep Learning Hybrid Model for Workers Activity Recognition

In various sectors such as elder care, disability support, athletics, and notably construction, The demand for accurate human activity recognition is increasing. In the construction industry, accurately classifying workers' activities is crucial for improving safety, productivity, and project management. This involves aspects like hazard identification, accident prevention, progress tracking, resource allocation, quality control, legal compliance, and worker assessment. Accurate activity classification is also key to advancing autonomous systems in construction. This work highlights the importance of a Hybrid Model using Deep Learning that combines a 1D-Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM). This CNN-LSTM model effectively categorizes Construction Workers' Activities (CWA) using accelerometer data from smartphones. The hybrid model utilizes both spatial and temporal data components, exhibiting enhanced performance in comparison with individual CNN and LSTM models. Comparative evaluations show the CNN-LSTM model achieves a significant average 99.20% accuracy on an internally recorded data set of Construction Workers’ Activities and 93.54% accuracy on the UCI-HAR public dataset. This research underscores the model's effectiveness in enhancing activity recognition within the construction industry