Computer Vision for PPE Compliance Detection on the Factory Floor

A steel service center in Indiana installed a computer vision PPE compliance system at the entrances to their warehouse floor and near their slitting and shearing lines. In the first month, the system detected 847 PPE violations: 312 instances of missing safety glasses, 289 instances of missing hard hats, and 246 instances of missing high-visibility vests. The safety manager had estimated, based on periodic spot checks, that they had maybe 50 PPE violations per month. The actual number was 17 times higher.

Within 3 months of the system going live (with real-time alerts to area supervisors), violations dropped to 94 per month. At month 6, they were down to 41. The system didn't change any PPE requirements; it just made non-compliance visible.

How PPE Detection Works

PPE detection uses object detection models (typically YOLO-based architectures) trained on images of workers wearing and not wearing specific PPE items. The model processes video feeds from existing security cameras or purpose-installed cameras, detecting people in each frame and classifying whether they're wearing the required PPE for that area.

The detection pipeline runs in stages. First, a person detection model identifies each individual in the camera's field of view. Then, a PPE-specific model examines each detected person for the presence or absence of required items: hard hat, safety glasses, high-visibility vest, steel-toed boots (partially visible, so less reliable), gloves, face shields, hearing protection, and fall protection harnesses.

Accuracy varies by PPE type. Hard hats are the easiest to detect reliably (97% to 99% accuracy) because they have a distinctive silhouette from most camera angles. Safety glasses are harder (90% to 95%) because prescription glasses can be confused with safety glasses, and some safety glasses are nearly transparent and hard to distinguish from bare eyes at a distance. High-visibility vests are reliably detected (96% to 98%) due to their distinctive color and retroreflective stripes. Gloves and hearing protection are the most challenging (85% to 92%) because they're small relative to the body and often partially occluded.

Camera Placement and Lighting

The effectiveness of PPE detection depends heavily on camera placement. Cameras mounted high on walls or ceilings (the typical security camera position) see the tops of hard hats clearly but struggle with safety glasses and gloves. Cameras at eye level provide better detection of face-worn PPE but have more limited fields of view and are more likely to be blocked by equipment.

Most effective installations use a combination of camera angles. Entry/exit point cameras mounted at 5 to 6 feet height catch people walking through defined choke points, providing good visibility of all PPE types. Area monitoring cameras mounted at 10 to 12 feet height cover larger floor areas with reasonable detection accuracy for hard hats and vests, though reduced accuracy for smaller items.

Lighting matters more than most implementation guides suggest. A factory floor with bright overhead lighting and minimal shadows is ideal. Areas with high-contrast lighting (bright areas next to deep shadows), flashing warning lights, or welding arc light create challenging conditions for the vision system. IR-supplemented cameras can improve performance in low-light areas.

Alert Mechanisms and Response

The system can trigger alerts through several channels. Real-time audio alerts (a recorded voice message at the entry point saying "please put on your safety glasses") are effective for entry-point compliance. Text messages or push notifications to the area supervisor enable rapid response for violations detected on the production floor. Dashboard displays in the supervisor's office show current compliance status across all monitored areas.

The response workflow matters more than the detection technology. A manufacturing facility that generates alerts but doesn't act on them quickly teaches workers that the system has no consequences. The most effective implementations have a clear, documented response procedure: first violation gets a verbal reminder, second violation in the same week triggers a documented coaching conversation, and repeated violations involve formal corrective action.

Privacy considerations are real and should be addressed proactively. Most systems detect PPE compliance without identifying the specific individual (the system knows "a person without safety glasses is in Zone B" but doesn't know it's John). Some systems can identify individuals through badge detection or face recognition, but this raises privacy concerns that vary by jurisdiction and should be reviewed with legal counsel and employee representatives before implementation.

Beyond Detection: Behavioral Analysis

Advanced PPE compliance systems go beyond simple present/absent detection to analyze patterns. The AI can identify that PPE compliance drops during the last hour of each shift (fatigue effect), in areas near the break room (where workers remove PPE during breaks and forget to put it back on), and on specific days of the week (often Fridays in plants with Monday-through-Friday schedules).

These behavioral insights help safety managers target interventions more effectively. Instead of sending a generic reminder about PPE compliance, they can focus on the specific times, locations, and circumstances where violations are most likely. Some plants have used the data to redesign their PPE stations, placing safety glasses dispensers at the transition point between the break room and the production floor, for example, which reduced post-break violations by 60%.

Cost and Implementation

A basic PPE compliance system for a single entry point (camera, edge computing device, software) costs $5,000 to $15,000. A facility-wide system covering multiple zones with area monitoring cameras costs $40,000 to $120,000 depending on the number of cameras and the size of the facility. Annual software licensing runs $8,000 to $25,000.

The ROI calculation involves several factors. Direct savings from reduced OSHA citations (average serious violation penalty was over $16,000 in 2024) are quantifiable but relatively infrequent. The larger impact is on injury reduction: a study published in the Journal of Safety Research found that consistent PPE compliance reduces workplace injuries by 15% to 30%, depending on the industry and the types of PPE enforced. For a facility with annual worker's compensation costs of $200,000, a 20% reduction in injury-related costs saves $40,000 per year.

The cultural impact is perhaps the most significant long-term effect. When PPE compliance becomes a visible, measured, and managed aspect of daily operations rather than an occasional spot-check exercise, the overall safety culture shifts. Workers who consistently wear PPE because the system makes non-compliance impractical gradually develop the habit of wearing PPE because they understand its value. That transition from enforced compliance to habitual safety behavior is the ultimate goal, and visibility is the first step.