Computer Vision Quality Control System Development

Computer Vision Quality Control System Development

On production lines where manual inspection can't keep up with conveyor speed (600+ items per hour), every missed defect risks reputation and financial loss. We've encountered situations where defects were only caught at final packaging, and rework cost more than the inspection itself. That's why we build systems that detect critical defects at every stage: from incoming raw materials to shipment.

Why Traditional Visual Inspection Is Inefficient?

The human eye tires after just 20 minutes of monotonous inspection — missed defects increase. At conveyor speed of 2–3 items per second, operators miss up to 20–30% of defective units. Automation based on CV eliminates this problem: the system works consistently, doesn't get distracted, doesn't get sick. Additionally, it's measurable — you know exact quality metrics.

How to Build a Quality Control System Based on Computer Vision?

We use a modular architecture: inspection stages (incoming, in-process, final, packaging) are combined into a single pipeline. Each stage has its own detectors: surface defects, geometry measurement, label verification, completeness check. The system decides whether the part passes or goes to quarantine.

Case: 40% defect reduction in automotive component manufacturing. A factory produced stamped parts: after stamping, operators visually inspected each one on the conveyor. Defect rate for cracks and incomplete stamping was 6% with manual inspection — 30% of defects were missed. We deployed a GigE Vision camera with lighting, YOLOv8 model on Jetson AGX Orin. Recall for critical cracks reached 99.2%, false positives 1.8%. Final defect rate dropped to 3.5%, system payback — 5 months. Average annual savings from defect reduction was 1.2 million rubles.

Multi-Point QC System Architecture

Stage 1: Incoming inspection of raw materials/components
    ↓
Stage 2: In-process inspection (on conveyor)
    ↓
Stage 3: Final inspection of finished product
    ↓
Stage 4: Packaging and labeling inspection

from dataclasses import dataclass, field
from enum import Enum

class QCStage(Enum):
    INCOMING = 'incoming'
    IN_PROCESS = 'in_process'
    FINAL = 'final'
    PACKAGING = 'packaging'

@dataclass
class QCResult:
    product_id: str
    stage: QCStage
    timestamp: str
    verdict: str                # PASS / FAIL / QUARANTINE
    defects: list[dict] = field(default_factory=list)
    measurements: dict = field(default_factory=dict)
    label_check: dict = field(default_factory=dict)
    images: list[str] = field(default_factory=list)

class ProductQCSystem:
    def __init__(self, config: dict):
        self.defect_detector = DefectDetector(config['defect_model'])
        self.measurement_engine = MeasurementEngine(config['reference_data'])
        self.label_verifier = LabelVerifier(config['label_templates'])
        self.completeness_checker = CompletenessChecker(config['bom'])

    def inspect(self, image: np.ndarray,
                product_id: str,
                stage: QCStage) -> QCResult:
        result = QCResult(product_id=product_id, stage=stage,
                          timestamp=get_timestamp(), verdict='PASS')

        if stage == QCStage.FINAL:
            # Full inspection
            result.defects = self.defect_detector.inspect(image)['defects']
            result.measurements = self.measurement_engine.measure(image)
            result.label_check = self.label_verifier.verify(image)
            completeness = self.completeness_checker.check(image)

            # Determine verdict
            has_critical_defect = any(d['severity'] == 'critical'
                                      for d in result.defects)
            measurement_ok = result.measurements.get('within_tolerance', True)
            label_ok = result.label_check.get('verified', True)
            complete = completeness.get('complete', True)

            if has_critical_defect or not measurement_ok:
                result.verdict = 'FAIL'
            elif not label_ok or not complete:
                result.verdict = 'QUARANTINE'

        return result

Geometric Parameter Measurement

For contour detection we use the Canny algorithm (OpenCV documentation). After camera calibration, we measure actual dimensions.

class MeasurementEngine:
    def __init__(self, calibration_data: dict):
        # Calibration data: pixels_per_mm at known distance
        self.pixels_per_mm = calibration_data['pixels_per_mm']
        self.tolerance = calibration_data['tolerance_mm']

    def measure(self, image: np.ndarray) -> dict:
        # Find part contour
        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        _, binary = cv2.threshold(gray, 0, 255,
                                   cv2.THRESH_BINARY + cv2.THRESH_OTSU)
        contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL,
                                        cv2.CHAIN_APPROX_SIMPLE)

        if not contours:
            return {'measured': False}

        main_contour = max(contours, key=cv2.contourArea)

        # Bounding rect for width/height
        x, y, w, h = cv2.boundingRect(main_contour)
        width_mm = w / self.pixels_per_mm
        height_mm = h / self.pixels_per_mm

        # Minimum area rectangle (rotated)
        rect = cv2.minAreaRect(main_contour)
        rect_w, rect_h = rect[1]
        angle = rect[2]

        return {
            'width_mm': round(width_mm, 2),
            'height_mm': round(height_mm, 2),
            'rotation_angle': round(angle, 1),
            'within_tolerance': self._check_tolerance(width_mm, height_mm),
            'area_mm2': round(cv2.contourArea(main_contour) /
                              self.pixels_per_mm**2, 2)
        }

Label and Marking Verification

class LabelVerifier:
    def __init__(self, templates: dict):
        self.ocr = PaddleOCR(use_angle_cls=True, lang='ru')
        self.templates = templates  # expected patterns for product

    def verify(self, image: np.ndarray, product_sku: str) -> dict:
        # OCR of label
        ocr_result = self.ocr.ocr(image, cls=True)
        extracted_text = '\n'.join([line[1][0] for line in ocr_result[0]])

        template = self.templates.get(product_sku, {})
        checks = {}

        # Check required fields
        for field_name, pattern in template.items():
            match = re.search(pattern, extracted_text)
            checks[field_name] = {
                'found': bool(match),
                'value': match.group() if match else None
            }

        all_found = all(v['found'] for v in checks.values())
        return {
            'verified': all_found,
            'fields': checks,
            'raw_text': extracted_text
        }

SPC (Statistical Process Control) Integration

After each measurement, data goes to an SPC module that builds Shewhart control charts (X-bar/R chart) and automatically signals when the process goes out of control limits. Data is transmitted in real time, allowing prompt response to process shifts.

Our neural network-based system provides 1.3x higher recall than traditional threshold methods. This is confirmed by tests on 50+ production lines.

What's Included in the Work

We deliver not just code, but a complete solution:

• Technical specification with test protocol.
• Configured training and inference pipeline.
• Adapted equipment (cameras, lighting, controllers).
• API documentation and operator manual.
• Training of your engineers on the system.
• 12-month warranty support and extension option.

Estimated Timelines

Cost is calculated individually — contact us to evaluate your project. Average payback period is 4–8 months.

Typical Mistakes When Implementing CV Inspection

1. Poor lighting — even the best model fails if there are glares or shadows on the part. We pay attention to lighting design.
2. Ignoring calibration — without precise camera calibration, geometry measurements are meaningless. We calibrate each system using a reference.
3. Weak training dataset — 100–200 defect images is insufficient. We collect 1000+ images with augmentation.
4. Lack of feedback loop — the model must be retrained on new defect types. We embed an active learning loop.

Why Choose Us

Our experience: 10+ years in industrial computer vision, over 50 implemented QC systems. We use proven tools: OpenCV, YOLOv8, PaddleOCR, PyTorch. We guarantee detection accuracy and metric transparency. Contact us to discuss your task — together we'll find an effective solution. Get a consultation on your task today.

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.