Abstract: If you’ve ever spent even a few minutes at an active construction site, you will know it’s rarely quiet or predictable.There’s constant movement materials being shifted, machines running, people coordinating tasks, and often someone working several feet above the ground. In that kind of setting, even a minor lapse in attention can lead to a serious incident. Most accidents don’t happen because the work itself is impossible. More often, they occur because basic precautions like wearing Personal Protective Equipment (PPE) — are ignored or treated casually.
On most sites, safety is mainly handled by supervisors who walk around and keep an eye on things, or sit and watch CCTV footage. It definitely helps, but it’s not flawless. After all, no person can keep track of several workers and multiple camera screens for hours without feeling tired or losing focus. Over time, small violations can slip through simply because human attention isn’t unlimited. As construction projects grow larger and more complex, this gap becomes more noticeable.
This project looks at the problem from a slightly different angle. Instead of entirely relying on manual oversight, it introduces an automated safety monitoring system based on computer vision and deep learning. The system analyses live video streams from site cameras to detect workers and verify whether required PPE is being worn. Once a worker is identified, an object tracking module continues to follow that individual across frames. In other words, the system doesn’t just detect once and move on it keeps observing, which helps avoid missing unsafe actions during movement.
Seeing a worker in the frame isn’t enough to judge their behaviour. So, pose estimation helps interpret their body position and movement during work.For example,if someone climbs without stable support or works at height without a harness, the system can interpret that posture as potentially unsafe. A rule-based alert mechanism further checks predefined safety conditions, such as entering restricted areas or remaining without protective equipment beyond an acceptable duration. When a violation is identified, the system generates a visible warning and notifies the supervisor. The intention is not to replace human decision- making, but to strengthen it and ensure quicker responses when needed.
By combining PPE detection, worker tracking, behaviour analysis, and real-time alerts within a single framework, the system improves overall site monitoring. It eases the burden on supervisors, supports early risk detection, and encourages more consistent compliance with safety standards. Ultimately, the goal is straightforward: reduce preventable accidents and create a work environment where safety is actively supported rather than assumed.

Keywords: Computer vision, deep learning, personal protective equipment (PPE), safety monitoring, YOLO, CNN, OpenCV, object tracking, pose estimation, real-time alerting.

Downloads: | DOI: 10.17148/IJIREEICE.2026.14389