Computer Vision Damage Detection (Cracks, Dents, Scratches)

Recently, an insurance company approached us: they needed to automatically assess body damage from smartphone photos. The main pain point was tiny scratches and dents that operators missed. We developed a model that detects defects as small as 0.5 mm and classifies them by severity. A localization error could cost millions, so we aim for recall >95% with controlled false positives. Over 5 years in the market, we have completed more than 50 computer vision projects, including defect detection for automotive and metallurgy industries. Manual labor savings reach 80%, and claims drop by 30%. Get a preliminary estimate — contact us for a consultation.

Types of Damage and Detection Specifics

Cracks — thin linear structures with a small width-to-length ratio. Standard detectors perform poorly: bounding boxes are large while the defect is small. Segmentation is preferable.

Dents — surface deformation without material rupture. Hard to detect in 2D; raking light and 3D reconstruction help.

Scratches — similar to cracks, linear structures. Depth affects severity.

How We Detect Fine Cracks and Scratches

We use a combination of YOLOv8 for instance segmentation and specialized preprocessing. Raking light at a shallow angle casts shadows from the tiniest irregularities, revealing defects only a few pixels wide. The pipeline includes top-hat transformations and contrast enhancement.

System Architecture

from ultralytics import YOLO
import numpy as np
import cv2

class DamageDetectionSystem:
    def __init__(self, config: dict):
        # Детектор повреждений (YOLOv8 instance segmentation)
        self.detector = YOLO(config['detection_model'])

        # Классификатор тяжести
        self.severity_classifier = load_severity_model(config['severity_model'])

        # Измеритель размеров (требует калибровки)
        self.pixels_per_mm = config.get('pixels_per_mm')

    def analyze(self, image: np.ndarray) -> dict:
        # Детекция и сегментация повреждений
        results = self.detector(image, conf=0.4, iou=0.5)

        damages = []
        for i, (box, mask) in enumerate(zip(
            results[0].boxes,
            results[0].masks.data if results[0].masks else []
        )):
            damage_type = self.detector.model.names[int(box.cls)]
            bbox = box.xyxy[0].tolist()
            area_px = int(mask.sum().item())

            # Вырезаем регион для классификации тяжести
            x1, y1, x2, y2 = map(int, bbox)
            crop = image[y1:y2, x1:x2]
            severity = self.severity_classifier.predict(crop)

            # Реальные размеры если есть калибровка
            size_info = {}
            if self.pixels_per_mm:
                size_info['area_mm2'] = round(area_px / self.pixels_per_mm**2, 2)
                size_info['length_mm'] = self._estimate_length(mask)

            damages.append({
                'id': i,
                'type': damage_type,
                'severity': severity,
                'bbox': bbox,
                'area_pixels': area_px,
                'confidence': float(box.conf),
                **size_info
            })

        return {
            'damages': damages,
            'total_count': len(damages),
            'has_critical': any(d['severity'] == 'critical' for d in damages),
            'summary': self._generate_summary(damages)
        }

Why Raking Light Is Effective

For cracks and scratches, standard frontal lighting is insufficient. Raking light — a source at a shallow angle to the surface — casts shadows from the tiniest irregularities. We use top-hat transformation to extract fine details and histogram equalization to enhance contrast. This increases recall for small defects by 15-20%.

def process_raking_light_image(image_path: str) -> np.ndarray:
    """Normalized image with raking light"""
    # При правильном освещении на стенде — дополнительная обработка:
    img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)

    # Топ-hat трансформация для выделения мелких деталей
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (25, 25))
    tophat = cv2.morphologyEx(img, cv2.MORPH_TOPHAT, kernel)

    # Усиление контраста
    enhanced = cv2.equalizeHist(tophat)

    return enhanced

How We Measure Defect Sizes

After detection, we need to estimate real dimensions — crack length or dent area. We calibrate the camera with a reference object. The parameter pixels_per_mm is stored in the system config. Then, using the defect mask, we compute physical sizes. For cracks, we skeletonize the mask for more accurate length.

def measure_crack_length(mask: np.ndarray,
                          pixels_per_mm: float) -> float:
    """Measure crack length from mask skeleton"""
    from skimage.morphology import skeletonize

    skeleton = skeletonize(mask > 0)
    length_px = skeleton.sum()
    return round(length_px / pixels_per_mm, 2)

For precise measurements, calibration using a reference is required. We use a chessboard with known spacing, determine the camera matrix, and derive the pixels_per_mm factor.

Datasets and Metrics

Public datasets:

• NEU Surface Defect — 6 classes of steel defects, 1800 images
• DAGM — texture defects, 10 categories
• AITEX Fabric — fabric defects
• Concrete Crack Images — cracks in concrete

# Example training on NEU Surface Defect
from ultralytics import YOLO

model = YOLO('yolov8m-seg.pt')
model.train(
    data='neu_defect.yaml',
    epochs=150,
    imgsz=640,
    batch=16,
    workers=8,
    optimizer='AdamW',
    lr0=5e-4,
    augment=True,
    degrees=180,         # дефекты могут быть в любой ориентации
    fliplr=0.5,
    flipud=0.5,
    mosaic=1.0
)

Metrics on Different Materials

What's Included in the Work

1. Requirements analysis and dataset collection (shoot defects on your equipment or use ready-made datasets)
2. Development of detection and segmentation model, training with augmentation
3. Integration into your IT infrastructure (REST API, Docker container)
4. Camera calibration and lighting setup (raking light rig if needed)
5. Documentation and operator training
6. 3-month warranty support

The cost of developing such a system depends on complexity and data volume. We evaluate each project individually to define the scope. Get a preliminary estimate — contact us.

Implementation Timeline

Checklist: Production Readiness for AI Defect Detection

Before starting the project, check the following:

• Controlled lighting on the stand (raking or diffuse light).
• Stable camera mount with resolution at least 5 MP for defects from 0.5 mm.
• Historical image archive: at least 200–300 examples of each defect type.
• Clear acceptance criteria: maximum allowed false positive and miss rates.
• Readiness for data labeling (mask annotation) or a pre-labeled dataset.
• Defined SLA: acceptable inspection time per object.

We guarantee transparent support at all stages. Request an engineer consultation — discuss the details of your project. Get a solution that really works.

NEU Surface Defect dataset (Song et al., 2019)

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.