Real-time Hazard Detection System for Construction Safety using Hybrid YOLO-ViTs

Use an AI-enabled tool in order to improve safety on a construction site by leveraging a hybrid YOLO-ViTs object detection model. The system will detect a worker who is not wearing the helmet, a worker without having the safety vest, a worker near the machinery or vehicles, and a worker entering in restricted areas. The restricted areas are calculated dynamically by using the clustering of the coordinates of the safety cone. Real-time hazard detection through video stream analysis will be made with prompt alerts and notifications through the integration services of LINE, Messenger, WeChat, and Telegram. Further, the post-processing algorithms also improve the detection accuracy. Moreover, it supports multiple languages, so it is useful to users worldwide. It shall also be configurable through YAML, hence easy deployment with real-time integration via a web interface is possible. The majority of the dataset comes from Roboflow, and the labeling is specific to construction. The roadmap of development includes real-time deployment on actual construction sites and is in ongoing updates and maintenance according to user feedback. Accessibility of the system to varied messaging platforms and, further, the multi-language notification feature make it highly versatile to use in application development for ensuring safety on-site across the global globe.