AI-Powered Safety Monitoring for Construction Sites

Cameras are present on every construction site, but they record archives that nobody watches. An inspector visits once a week—and not always. The result: violations are only recorded after an incident. Our computer vision system works 24/7: it detects missing hard hats, vests, presence in dangerous zones and sends an alert within seconds. Unlike a human, AI does not get tired or distracted. We measured: on a site with 12 cameras, the system detects 5 times more violations than manual patrols. According to Rostechnadzor data, 80% of safety violations on construction sites are related to missing PPE. Our technology bridges the gap between periodic inspections and continuous monitoring.

What Violations Does the System Detect?

The system detects up to 7 types of violations in real time:

How We Build a PPE Detector

import cv2
import numpy as np
from ultralytics import YOLO
from dataclasses import dataclass

@dataclass
class SafetyViolation:
    violation_type: str
    worker_id: int
    bbox: list
    confidence: float
    zone: str
    severity: str  # 'warning', 'critical'

class ConstructionSafetyMonitor:
    def __init__(self, model_path: str, config: dict):
        # YOLOv8l fine-tuned on Safety Helmet Dataset + custom PPE data
        # Classes: person, hard_hat, safety_vest, no_hard_hat, no_vest,
        #         safety_glasses, gloves, harness
        self.model = YOLO(model_path)

        self.danger_zones = config['danger_zones']
        self.required_ppe = config.get('required_ppe',
                                        ['hard_hat', 'safety_vest'])
        self.violation_history = {}  # worker_track_id -> violations

        # Additional pose estimator for harness check at height
        self.pose_estimator = YOLO('yolov8l-pose.pt')

    def _worker_has_ppe(self, worker_bbox: list,
                         ppe_detections: list,
                         ppe_class: str) -> tuple[bool, float]:
        """Check if a specific worker has the required PPE"""
        wx1, wy1, wx2, wy2 = worker_bbox
        worker_upper_half = [wx1, wy1, wx2, wy1 + (wy2 - wy1) * 0.6]

        best_iou = 0.0
        for ppe in ppe_detections:
            if ppe['class'] == ppe_class:
                iou = self._iou(worker_upper_half, ppe['bbox'])
                best_iou = max(best_iou, iou)

        # IoU > 0.1 = PPE is located in the worker's body area
        return best_iou > 0.1, best_iou

    def detect_violations(self, frame: np.ndarray) -> list[SafetyViolation]:
        results = self.model.track(frame, persist=True, conf=0.4)
        violations = []

        persons = []
        ppe_items = []

        for box in results[0].boxes:
            cls = self.model.names[int(box.cls)]
            bbox = list(map(int, box.xyxy[0]))
            conf = float(box.conf)
            track_id = int(box.id) if box.id is not None else -1

            if cls == 'person':
                persons.append({'bbox': bbox, 'track_id': track_id})
            elif cls in ['hard_hat', 'safety_vest', 'safety_glasses',
                          'gloves', 'harness']:
                ppe_items.append({'class': cls, 'bbox': bbox, 'conf': conf})

        # For each worker, check PPE presence
        for worker in persons:
            zone = self._get_zone(worker['bbox'])

            for required in self.required_ppe:
                has_ppe, iou_score = self._worker_has_ppe(
                    worker['bbox'], ppe_items, required
                )

                if not has_ppe:
                    vtype = f'no_{required}'
                    violations.append(SafetyViolation(
                        violation_type=vtype,
                        worker_id=worker['track_id'],
                        bbox=worker['bbox'],
                        confidence=1.0 - iou_score,
                        zone=zone,
                        severity='critical' if required == 'hard_hat' else 'warning'
                    ))

            # Check restricted zone entry
            if zone in self.danger_zones:
                cx = (worker['bbox'][0] + worker['bbox'][2]) // 2
                cy = (worker['bbox'][1] + worker['bbox'][3]) // 2
                if self._in_polygon(cx, cy,
                                     self.danger_zones[zone]['polygon']):
                    violations.append(SafetyViolation(
                        violation_type='unauthorized_zone_entry',
                        worker_id=worker['track_id'],
                        bbox=worker['bbox'],
                        confidence=0.95,
                        zone=zone,
                        severity='critical'
                    ))

        return violations

    def _iou(self, box1: list, box2: list) -> float:
        x1 = max(box1[0], box2[0])
        y1 = max(box1[1], box2[1])
        x2 = min(box1[2], box2[2])
        y2 = min(box1[3], box2[3])

        inter = max(0, x2-x1) * max(0, y2-y1)
        area1 = (box1[2]-box1[0]) * (box1[3]-box1[1])
        area2 = (box2[2]-box2[0]) * (box2[3]-box2[1])
        union = area1 + area2 - inter
        return inter / max(union, 1e-6)

Case Study: Residential Complex with 200 Workers

On site, we installed 12 IP cameras. Before deployment, an inspector walked the site once a day and recorded violations manually. The average number of PPE violations was 30–40 per day, some going unnoticed.

After system deployment:

• Coverage: 100% of camera vision zones in real time
• Violations detected in first week: 847 (vs 40–50 manually)
• After one month of operation: 73% reduction in violations
• 2 critical incidents prevented (presence in crane operation zone)

Accuracy on test set: 91% for hard hats, 87% for vests (challenging case—vest worn but unbuttoned).

Why Choose Our Safety System?

We use a fine-tuned YOLOv8 model—the best balance of speed and accuracy. Inference on NVIDIA T4 takes 8–10 ms per frame, enabling 30 FPS processing without lag. The system uses augmentation and hard negative mining techniques for robustness to shadows and weather. We guarantee accuracy no lower than 85% on major PPE classes in industrial operation. Preventing one critical incident can save up to $60,000 in fines and downtime. Average savings from deploying the system on a site with 10+ cameras amount to $25,000–$35,000 per year.

What's Included in Turnkey Work

• Site audit and camera selection
• Software installation on your server or cloud
• Model fine-tuning for site specifics
• Integration with NVR and alerting system
• Training for safety personnel on dashboard use
• 2 months of technical support after launch

How Does the System Integrate with Video Surveillance?

The system connects to NVR or directly to cameras via RTSP. Popular brands supported: Hikvision, Dahua, Uniview. The API can send alerts to any safety management system (EAM, SCADA). For a pilot project, 2–4 cameras with 1080p resolution and a server with GPU are sufficient. Recommended server configuration: GPU NVIDIA T4 (16GB VRAM), 16 GB RAM, 500 GB SSD. For edge deployment—NVIDIA Jetson Orin NX. Cloud GPU instances can be used.

Deployment in 4–12 Weeks

1. Analysis (1 week): Site visit, coverage assessment, KPI agreement.
2. Design (1–2 weeks): Camera and server selection, RTSP stream setup.
3. Implementation (2–4 weeks): Model training, inference pipeline preparation, alert writing.
4. Testing (1 week): Run on historical recordings, A/B test in pilot zone.
5. Deployment (1 week): Production installation, staff training, documentation handover.

Request a demo on your actual construction site data. Get consultation on equipment selection and metrics.

Notifications and Integration

• Guard alerts via Telegram bot with violation photo
• Automatic violation report generation with frame, time, zone
• Violation statistics export to Excel/Power BI for safety manager

We guarantee stable operation in all weather conditions. Contact us for a demo on your actual data. Our engineers hold NVIDIA Jetson certifications and have deployment experience on sites with 50+ cameras.

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.