Construction workers' unsafe behavior detection through adaptive spatiotemporal sampling and optimized attention - video monitoring

Oct 22, 2026

Uploaded by Angel

Construction site image analysis faces difficulties in accurate object detection and identification of unsafe actions due to complex backgrounds, varying object sizes, and inconsistent image quality. Existing attention-based approaches often miss adaptive sampling and channel-wise feature adjustments, overlooking potential spatiotemporal redundancies. This paper presents OP-Net architectures and an attention-based spatiotemporal sampling strategy, improving object detection and enabling effective unsafe action identification. Experiments evaluate detection on SODA and benchmark action identification on CMA with state-of-the-art accuracy and reasonable efficiency.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Introduction

Keywords and Abstract

Automation in Construction 165 (2024) 105508 Contents lists available at ScienceDirect Automation in Construction journal [homepage:](homepage: www.elsevier.com/locate/autcon)[ www.elsevier.com/locate/autcon](homepage: www.elsevier.com/locate/autcon) | Construction workers’ unsafe behavior detection through adaptive spatiotemporal sampling and optimized attention based video monitoring Mahdi Bonyani , Maryam Soleymani , Chao Wang *
Bert S. Turner Department of Construction Management, Louisiana State University, USA | | | | | --- | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | | Keywords:
Unsafe behavior detection Construction worker safety Adaptive spatiotemporal sampling Attention learning
Video analysis
Deep learning | | In recent years, advances in construction site image analysis faced challenges, particularly in construction object detection and identifying unsafe actions. Challenges involve complex backgrounds, varying object sizes, and image quality. Existing methods address spatial and temporal features with attention mechanisms but often overlook adaptive sampling and channel-wise adjustments, missing potential spatiotemporal redundancies. This article introduces the Optimized Positioning (OP-Net) architectures and an attention-based spatiotemporal sampling approach. The OP module is introduced for object detection, which enhances channel relationships by leveraging global feature affinity associations. Additionally, we propose an innovative spatiotemporal sampling strategy that adapts to effectively identify unsafe actions in construction sites. We extensively evaluate the object detection task using the SODA dataset to showcase the efficacy and effectiveness of our approach. Furthermore, our unsafe action identification model is benchmarked on the CMA dataset, demonstrating its ability to achieve new state-of-the-art performance in accuracy while maintaining reasonable computational efficiency. | | 1. Introduction Construction and civil infrastructure projects often entail complex interactions between individuals and machinery, each involving unique and dynamic processes [1]. Safety in the construction industry has been a longstanding global concern, with significantly higher rates of injuries and fatalities compared to other sectors [2]. The importance of ensuring safety at construction sites has never been more critical, especially as the industry experiences revitalization and an increased demand for infrastructure development. To illustrate, in the UK, the construction sector accounts for just 5% of the workforce but is responsible for 27% of fatal injuries, while in the US, construction accounted for nearly 20% of all occupational fatalities in 2016–2017 [3]. Despite extensive efforts to enhance safety through measures like Occupational Safety and Health Administration (OSHA) requirements, accident and mortality rates in construction have remained persistently high or reached a plateau in recent years [4]. Studies indicate that a substantial portion of construction accidents can be attributed to human factors such as unsafe behavior, lapses in supervision, and a lack of risk awareness [5]. Cognitive factors play a significant role in determining the ability of workers to promptly recognize and respond to hazards in the everchanging environment of construction sites. Although advanced technologies like virtual reality have been explored to enhance risk awareness, they may not fully address the fundamental cognitive limitations that impact visual search and risk assessment [6]. Research indicates that a significant proportion of construction accidents, ranging from 70% to 88%, can be attributed to workers’ unsafe behaviors as the immediate cause [5,7,8]. Studies conducted on construction sites have identified elevated rates of unsafe actions, such asthe failure to use fall protection equipment, in various countries, including the US, UK, and China. Despite eff

FAQs

What challenges does construction site image analysis face for unsafe action detection?

It struggles with complex backgrounds, varying object sizes, and image quality that complicate accurate object detection and unsafe action identification.

What does the paper propose to address these challenges?

It introduces OP-Net architectures and an attention-based spatiotemporal sampling approach, including an OP module to enhance channel relationships via global feature affinity.

How is the proposed method evaluated and what datasets are used?

Object detection is evaluated on the SODA dataset, while unsafe action identification is benchmarked on the CMA dataset, achieving new state-of-the-art accuracy with reasonable computational efficiency.

Oct 22, 2026

Uploaded by Angel