MULTIMODAL ARTIFICIAL INTELLIGENCE FOR REAL-TIME HAZARD DETECTION AND RISK PREDICTION IN DYNAMIC OCCUPATIONAL ENVIRONMENTS: A SYSTEMATIC REVIEW AND CONCEPTUAL FRAMEWORK

Abstract

Occupational hazards in dynamic environments such as construction, mining, manufacturing, and oil and gas continue to cause significant injuries and fatalities despite advances in safety management. This study presents a systematic review of sources published between 2022 and 2026, examining the application of multimodal artificial intelligence for real-time hazard detection and risk prediction. The review reveals a rapid shift from unimodal systems (primarily computer vision and wearable sensors) toward integrated multimodal approaches. Key findings show that multimodal fusion architectures achieve 15–30% higher accuracy in dynamic conditions compared to unimodal baselines, with vision-language models demonstrating 85–92% performance in contextual hazard reasoning. However, persistent challenges include occlusion and lighting issues, data scarcity, latency in complex fusion, limited real-world validation, and ethical concerns surrounding continuous worker monitoring. To address these gaps, this paper proposes the Dynamic Occupational Multimodal AI Safety (DOMAS) Framework— a novel four-layer architecture that integrates heterogeneous data acquisition, adaptive attention-based fusion, edge-deployed real-time inference, and explainable closed-loop intervention. The framework is designed for sub-500 ms latency on edge devices and targets a projected 25–40% reduction in occupational incidents based on observed multimodal gains. This work advances multimodal AI theory in safety-critical domains and provides industry, policymakers, and developers with a practical blueprint for deployment. By bridging fragmented research and offering a unified, human-centric solution, the DOMAS Framework contributes meaningfully to the goals of Industry 5.0 and zero-harm workplaces. Future research priorities include large-scale multimodal datasets, longitudinal field trials, and ethical governance frameworks.