Development and Evaluation of an AR-Based Interactive System for Occupational Safety Education

In the field of labor-intensive production, occupational safety represents a fundamental prerequisite for safeguarding workers’ health and ensuring the stable development of enterprises. With increasing industrial complexity and the proliferation of safety hazards, traditional approaches to occupational safety education—predominantly reliant on passive knowledge transmission—have proven insufficient in facilitating mastery of safety protocols, frequently resulting in non-compliant behavior during practical operations. Augmented reality (AR) technology has emerged as a promising solution, offering immersive learning environments for safety training. However, limitations persist in existing AR-based safety education systems. Specifically, the use of complex backbone networks in recognition models has resulted in slow running speed and difficulty in meeting real-time interaction requirements. The accuracy of compliance assessment remains suboptimal due to inadequate consideration of scene-specific variations. Furthermore, limited feature extraction capabilities and insufficient attention mechanisms have hindered the system’s ability to distinguish between similar operations, thereby compromising recognition precision. Therefore, this study focuses on the development of a compliance assessment method for user-interactive operations based on AR gesture recognition. A lightweight network architecture, ShuffleNetv2, was introduced to replace the original backbone network, thereby reducing computational complexity and enhancing operational efficiency. Additionally, a dynamic selective attention mechanism, SKNet, was integrated into the model to enhance the extraction of critical operational features, thereby improving the accuracy of compliance determination to address limitations identified in prior research.