Recognizing sitting activities of excavator operators using multi-sensor data fusion with machine learning and deep learning algorithms

Abstract

Recognizing excavator operators' sitting activities is crucial for improving their health, safety, and productivity. Moreover, it provides essential information for comprehending operators' behavior patterns and their interaction with construction equipment. However, limited research has been conducted on recognizing excavator operators' sitting activities. This paper presents a method for recognizing excavator operators' sitting activities by leveraging multi-sensor data and employing machine learning and deep learning algorithms. A multi-sensor system integrating interface pressure sensor arrays and inertial measurement units was developed to capture excavator operators' sitting activity information at a real construction site. Results suggest that the gated recurrent unit achieved outstanding performance, with 98.50% accuracy for static sitting postures and 94.25% accuracy for compound sitting actions. Moreover, several multi-sensor combination schemes were proposed to strike a balance between practicability and recognition accuracy. These findings demonstrate the feasibility and potential of the proposed approach for recognizing operators' sitting activities on construction sites.