Personalized stability monitoring based on body postures of construction workers working at heights

• Published in: Safety science Vol. 162; p. 106104
• Main Authors: Duan, Pinsheng, Goh, Yang Miang, Zhou, Jianliang
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2023
• Subjects: Construction workers; Gaussian mixture model; Posture recognition; Stability monitoring; Working at height; Posture recognition; Working at height; Gaussian mixture model; Stability monitoring; Construction workers
• ISSN: 0925-7535; 1879-1042
• DOI: 10.1016/j.ssci.2023.106104

•A vision-based stability monitoring method based on body postures.•This study helps to monitor workers at heights who are physical instability.•A worker working on suspend scaffolding was used to validate the approach.•Proposed method can help personalized and adaptive behaviour-based safety training. Fall from heights accidents are one of the most frequent causes of death in the construction industry. Many fall-from-height accidents are related to stability issues and stability is critical for construction workers working at heights. Thus, stability monitoring of workers is important for proactive accident prevention. However, workers’ stability is highly personalized due to differences in workers’ physical characteristics and habits. Little attention has been paid to the personalized posture-based stability analysis of workers working at heights. This study proposes a personalized stability monitoring framework based on workers' body posture patterns when working at heights. The proposed method includes two main components: posture recognition and stability monitoring. First, OpenPose is used to extract the coordinates of workers’ posture key points from video clips. Two posture features, duration and count, are extracted to reflect workload and frequency. Secondly, a two-stage stability monitoring framework, including instability detection and instability evaluation, is designed based on the Gaussian model and Gaussian mixture model. Finally, the proposed framework is validated by on-site construction videos of workers working at height. The validation results showed the accuracy of 84.38% and the precision of 81.25% in identifying the stable status of a subject. The robustness of the personalised monitoring method was validated through comparisons with five other workers. The study provides a practical reference for active safety monitoring for workers with high fall-from-height risk. It also helps to extend personalized and adaptive behaviour-based safety training for construction workers.