Leveraging OWL-ViT and MQTT for Real-Time Construction Safety

In Hong Kong, many construction assistants die every year in various non-fatal construction accidents because they do not wear construction site helmets. Traditional manual supervision often has problems of low efficiency, negligence, and excessive capital costs. To address these issues, this paper proposes an intelligent monitoring system based on AloT that integrates deep learning and real-time data communication to enhance safety compliance and environmental oversight. The system uses the OWL-ViT model for automatic helmet detection and uses MQTT for seamless data transfer between the client and server. Key features include checking worker helmet-wearing rationality, temperature, and humidity, and escalating decibel tracking. Experiments show that the system can provide operability opinions for site safety. This study highlights the system's problematic potential to reduce accidents on construction sites and improve the built environment.