An Integrated Artificial Intelligence Framework for Enhancing Safety Monitoring and Identifying Workers' Safety Violations in the Dynamic Construction Environment

The notable fatality rate within the construction industry presents pressing challenges for both industry practitioners and scholars. A paucity of safety officers hampers rigorous site monitoring and risk detection. With the advent of artificial intelligence (AI), there's a marked improvement in the automated identification of safety violations, offsetting the shortfall of safety managers. However, despite the promise of automation, the mutable environment of construction sites challenges its effectiveness and scalability.

Addressing this safety challenge of intricate construction sites, this research developed an integrated framework that leverages AI to identify worker-related safety violations in dynamic construction settings. The framework consists of the following four interrelated systems: As the foundational system of the framework, the Camera Placement Optimization System serves a crucial purpose. Firstly, it determines the optimal placement of surveillance cameras, considering the construction schedule, to ensure maximum coverage of the construction site and minimize potential blind spots. Secondly, after optimal positioning, these cameras initiate the critical process of capturing exhaustive visual data on worker activities. This visual information is invaluable to the subsequent systems in the framework. Drawing on the visual data procured by the cameras positioned by the Camera Placement Optimization System, the second system, the Struck-by Hazard Detection System, identifies potential hazards where workers could be struck by construction equipment. By using stereo-vision techniques, this system provides a detailed assessment of such risks within the ever-evolving construction environment. The third system, acting as a nexus, is the Data Integration System, which amalgamates several distinct categories of data identified by multiple computer vision methods. In this framework, with its emphasis on creating comprehensive data, it refines and structures the visual data into a unified format, ensuring integrated information in the final safety identification phase. As the final step of the integrated framework, building on the consolidated data from the Data Integration System, the Safety Violation Identification System adopted a graph-typed knowledge base and advanced AI algorithms to identify workers' safety violations. In the knowledge base, it employs engineering-related knowledge. Specifically, it utilizes safety rules defining workers' safety behavior and incorporates public engineering standards and best practices, such as the hardhat classification on construction sites. This system accurately pinpoints the safety violations and ensures the unsafe behaviors can be interrupted efficiently. In addition, to further enhance the detection efficiency and scope of the integrated framework, we established a specialized image dataset centered on construction equipment. This dataset empowers the AI with refined detection capabilities, thereby broadening the detectability scope and strengthening the overall surveillance extent within the construction environments.

The framework introduced in this research demonstrates commendable resilience and efficiency across diverse construction environments. Distinctively, its inherent modularity and scalability ensure adaptability, satisfying the specific demands of individual construction. Through the integration of multiple systems, this framework is to mitigate risks in diverse construction environments throughout the construction lifecycle. The automation in surveillance and monitoring acts as an assistance, leading workers leaves the hazardous situations, thus significantly reducing the likelihood of construction accidents. Moreover, the facility of real-time feedback equips project administrators with timely insights, bolstering data-centric decision-making processes. Incorporating a Building Information Modeling (BIM)-centric strategy for the optimization of camera placement not only optimizes spatial coverage but also contributes to cost-effective construction practices. In summary, by combining state-of-the-art AI with profound engineering knowledge, our framework emerges as an integrated solution, enhancing worker safety in the realm of construction.