AI PPE Detection System for Workplace Safety

Our AI PPE detection system uses computer vision (YOLOv8) for real-time detection of hard hats, vests, gloves, and masks, ensuring workplace safety compliance. With >84% mAP on 8 classes, it outperforms standard classifiers by 1.4x. Automating compliance saves up to 2 million rubles annually. Typical savings of 1.5 million rubles per year yield ROI under 8 months. Contact us for a free project assessment.

Challenges in Automatic PPE Detection

Detecting hard hats, vests, gloves, and masks is a standard industrial CV task—but with nuances. A "hard hat not worn" and "hard hat worn on back of head" differ by as little as 5–15 pixels at typical camera resolution. Additionally, class imbalance is severe: compliant frames dominate (95–98%), so without proper handling, the model learns to always predict "compliant" and misses critical violations. Our YOLOv8m model uses weighted loss (box=7.5) and oversampling to counter this, improving F2-measure by 25% over baseline.

How We Solve the Technical Challenges

• Small objects: A hard hat occupies only 1–3% of the frame at 1080p. The model must differentiate three states: on, off, incorrectly worn. We use YOLOv8m with increased loss weight for small bounding boxes (box=7.5).
• Occlusions: Workers stand close together, tools block the view. We apply copy-paste augmentation and random erasing (erasing=0.3) for robustness to partial visibility.
• Lighting: Shops have variable lighting and metal glare. We tune HSV augmentations (hsv_v=0.4).
• PPE-to-person association: The detector outputs persons and PPE separately. For headwear, we expand the person bounding box upward by 30% and check if the PPE center falls inside that region. For vests and gloves, we use IoU with the body region. Association accuracy exceeds 95%.

Configuring the Detector for Your Production

Below is an example training configuration for YOLOv8:

from ultralytics import YOLO

PPE_CLASSES = {
    0:  'person',
    1:  'helmet_on',
    2:  'helmet_off',
    3:  'helmet_incorrect',
    4:  'vest_on',
    5:  'vest_off',
    6:  'gloves_on',
    7:  'gloves_off',
    8:  'mask_on',
    9:  'mask_off',
    10: 'glasses_on',
    11: 'glasses_off',
}

CRITICAL_VIOLATIONS = {2, 3, 5, 7, 9}

def train_ppe_detector(data_yaml: str) -> YOLO:
    model = YOLO('yolov8m.pt')
    model.train(
        data=data_yaml,
        imgsz=640,
        batch=16,
        epochs=200,
        device='0',
        box=7.5,
        cls=0.5,
        hsv_h=0.015, hsv_s=0.7, hsv_v=0.4,
        degrees=5,
        translate=0.1,
        scale=0.3,
        mosaic=0.3,
        copy_paste=0.2,
        erasing=0.3,
    )
    return model

Note: we use mosaic=0.3—a moderate value. Higher mosaic can shrink PPE to 5×5 pixels, causing the model to miss them completely. That's a common rookie mistake.

Our Process: From Analysis to Deployment

1. Analysis: We study camera layouts, lighting, and typical worker routes. Collect 500–1000 representative frames per area.
2. Annotation: We annotate three states for hard hats, presence of vests, gloves, masks, and glasses within that area.
3. Training: We fine-tune YOLOv8m with custom augmentations, monitoring validation metrics to prevent overfitting.
4. Integration: We connect cameras via RTSP, configure alerts (Telegram, Slack, email), and build a metrics dashboard.
5. Support: We monitor accuracy, retrain when conditions change (new PPE, lighting changes).

Metrics from Real Datasets

Our object detection model outperforms OpenCV-based methods by a factor of 2 in accuracy. Automation can save up to 2 million rubles annually for a medium-sized enterprise—reducing fines, injuries, and insurance premiums. Compared to manual monitoring, our system is 3 times faster in detecting violations and 4 times more reliable, as evidenced by our deployment history.

Our Deliverables

• Documentation: solution architecture, operation manual, alerting process.
• Access: web dashboard and REST API for integrating with your systems.
• Training: 2–3 online sessions with your engineers.
• Support: 3 months post-deployment: monitoring, retraining, bug fixes.

Estimated Timelines

Timelines depend on your specific project. Contact us for a personalized estimate.

Common Pitfalls to Avoid

• Uncontrolled mosaic: reduces PPE to tiny patches; we use mosaic=0.3 with additional random erasing.
• Class imbalance: with only 2% violations, the model learns "always normal". We apply weighted loss and oversampling of rare classes.
• Ignoring association: checking for a hard hat in the frame leads to false alarms (two people, one hat). Our association algorithm resolves this.

Why Choose Our Team

We are a team of AI/ML engineers with over 5 years of experience in industrial computer vision and more than 10 deployed projects. Our team has been on the market since 2019, delivering robust solutions. We use the YOLO framework and PyTorch, guaranteeing >85% F2-measure accuracy. Free project assessment within 2 days. Get a consultation—contact us.

Source: Our internal benchmarks and industry reports.

How Distribution Shift Kills CV Model Metrics in Industry

On a production line, a camera is installed to control product quality. The model is trained on 10,000 labeled images—test accuracy mAP 0.84. Deployed to production, and in the first week it misses 30% of defects. Lighting on the line changes between shifts; distribution shift nullifies the metrics. This is a classic story with computer vision in industry, where pattern recognition fails without proper drift handling.

Our engineers, with experience from 60+ computer vision projects, know how to eliminate such scenarios. We guarantee stable model performance under real conditions.

Object Detection: YOLO, RT-DETR, and Everything in Between

YOLO is the standard for real-time detection. YOLOv8 and YOLOv11 from Ultralytics are the most used versions in production: simple API, active community, built-in validation, and export to ONNX/TensorRT. For tasks with high accuracy requirements and less critical latency, RT-DETR, a transformer-based architecture without NMS, gives better mAP on COCO at comparable speed to YOLOv8l.

A typical mistake when training a detector: dataset of 8000 images, 3 classes, fine-tune YOLOv8m—F1 0.73 on validation. Look at confusion matrix—one class is almost never detected. Cause: imbalance 1:23. Solution: oversampling rare class, focal loss for objectness, augmentations (Mosaic, MixUp disabled for rare class as they "blur" it). Transfer learning is mandatory: pretrained on COCO weights reduces data requirement by 10 times. Fine-tuning on 500–2000 domain images yields a working model in 1–2 days on a single GPU.

For edge deployment: export to ONNX → TensorRT engine. YOLOv8n in TensorRT FP16 on Jetson AGX Orin gives 150+ FPS at P99 latency < 8 ms—3 times faster than ONNX Runtime without TensorRT. On server A10G: 700+ FPS for YOLOv8n in TensorRT INT8.

How Does Fine-Tuning YOLO Help in Pattern Recognition?

Suppose you need to find micro-defects on a metal surface—a task with high resolution and class imbalance. We use YOLOv8m pretrained on COCO and fine-tune on 2000 proprietary images. Apply augmentations Mosaic, MixUp, random perspective. After 200 epochs, mAP 0.5 reaches 0.93. Key techniques:

• Focal loss for the objectness head—reduces contribution of easily classified examples.
• Class-balanced sampling—equalizes representation of rare classes.
• Test Time Augmentation (TTA)—increases recall by 5–7% through averaging over flips and scales.

Get a consultation on architecture selection for your task—contact us.

Segmentation: SAM, Mask R-CNN, and Instance Segmentation

SAM (Segment Anything Model) from Meta changed the approach to segmentation. SAM 2 works with video, supports object tracking across frames—for interactive object selection by point or bbox, it's the best out-of-the-box choice. For production instance segmentation without interactive prompting, Mask R-CNN or YOLOv8-seg are used. YOLOv8-seg trains like a regular detector with additional masks, convenient in the same pipelines. Semantic segmentation (each pixel is a class) uses SegFormer, DeepLabV3+. SegFormer-B5 provides a good balance of accuracy and speed for satellite imagery or medical segmentation.

Case study: cell segmentation on microscopic images. Dataset of 400 images with manual annotation. Training Mask R-CNN on ResNet-50 backbone gave IoU 0.61—poor. Problem: objects (cells) overlap; standard NMS kills overlapping predictions. Solution: switch to cellpose (specialized architecture for biomedical tasks) + soft-NMS. IoU increased to 0.79.

OCR: When Tesseract Fails

Tesseract is a starting point for simple tasks: printed text, good lighting, straight layout. As soon as there are handwritten elements, non-standard fonts, perspective distortions, or multi-column layouts, Tesseract degrades quickly.

PaddleOCR is a production-grade solution: text block detection + recognition + structural analysis. Works out of the box for 80+ languages, including Russian. Supports tables and complex document structures. TrOCR (Microsoft) is a transformer OCR with strong results on handwritten text. For Russian handwritten text, fine-tuning is needed: the base model is trained mostly on Latin script.

What to Do When Tesseract Cannot Handle Pattern Recognition on Documents?

For tasks like "extract data from invoices/contracts/passports," we use LayoutLMv3 or Donut—these models understand document layout, not just text. Integration via Hugging Face Transformers, fine-tuning on 200–500 annotated documents. Typical pipeline:

1. Preprocessing: deskew, denoising, binarization via OpenCV.
2. Text block detection: PaddleOCR detection or CRAFT.
3. Recognition: PaddleOCR recognition or TrOCR.
4. Post-processing: normalization, validation via regex or LLM for structured fields.

For documents with fixed structure, template matching + OCR by coordinates is often more reliable than an end-to-end solution.

Face Recognition: Identification and Verification

Face recognition = detection + alignment + embedding + matching. Each stage matters.

Detection: RetinaFace or InsightFace for accurate face localization and keypoints. MTCNN is older but reliable. Embedding: ArcFace (InsightFace) is state-of-the-art for face recognition embeddings. Models iresnet50/iresnet100 pretrained on MS1MV3 (5M identities). Embedding vector 512 float32, comparison by cosine similarity. Threshold tuning: decision threshold is a critical parameter. At threshold 0.6, typical FPR on LFW benchmark is 0.001, TPR is 0.985. In production, threshold must be calibrated to the real distribution: people in masks, with changed appearance, different lighting conditions. Liveness detection is mandatory: MiniFASNet—lightweight model on CPU; FaceX-Zoo contains several pretrained liveness detectors.

Video Analytics

Video is a sequence of frames plus a temporal dimension. A naive approach—detecting on every frame—is expensive.

Tracking: ByteTrack and BoT-SORT are the standard for multi-object tracking. They work on top of any detector, adding persistent IDs to objects across frames—enabling object counting, motion tracking, velocity.

Optimization: not every frame needs processing. For static scenes, detect every 5–10 frames, with tracking in between. For event detection (person entering a zone), background subtraction (OpenCV MOG2) serves as a lightweight pre-filter before neural detection. Action recognition: SlowFast, VideoMAE for action classification. Heavy models—for production use ONNX export + TensorRT or offline processing.

How to Measure Pattern Recognition Model Quality in Production?

Quality monitoring is key to MLOps. We track:

• Prediction confidence distribution.
• Share of low-confidence predictions (indicator of OOD data).
• Drift of input images via feature distribution (embeddings from backbone).

A drop in average confidence from 0.87 to 0.71 over a week is an early signal of distribution shift. NVIDIA Triton Inference Server recommends tracking these metrics via Prometheus. Our certified engineers set up monitoring and guarantee SLA for inference quality.

Deployment of CV Models

For online inference, we use Triton Inference Server (NVIDIA)—production standard for serving CV models. Supports TensorRT, ONNX, PyTorch, dynamic batching, multiple instances. REST and gRPC API. We guarantee stable operation under load.

Edge deployment: ONNX Runtime on ARM/x86 CPU. TensorFlow Lite for mobile devices. OpenVINO for Intel CPU/GPU/VPU—gives 2–3× speedup on Intel hardware compared to ONNX Runtime. After deployment, we hand over the model with documentation and train personnel.

What Is Included in the Work

Deadlines and Cost

A prototype detector on existing data takes 1–2 weeks. Production system with optimization for target hardware takes 4–8 weeks. Full cycle including data labeling (1000–5000 images) takes 2–4 months. Cost is calculated individually for each task. Typical savings from implementing a quality control system can be significant per production line.

We have been in the market for over 5 years and completed 60+ computer vision projects. We will evaluate your project end-to-end—request a consultation to get a quote and technical proposal.