AI-Powered Road Pavement Inspection System

AI-Powered Road Pavement Inspection System

Potholes, cracks, ruts—annual repair costs run into billions. The problem is that by the time road workers visually detect a defect, it has already become critical. Traditional manual inspection takes weeks and relies on subjective judgment. We develop AI inspection using cameras mounted on vehicles or specialized machines. It detects early signs of pavement degradation and prioritizes repairs. Our certified engineers have over 7 years of experience in Computer Vision and have delivered more than 15 road inspection projects. Repair budget savings compared to manual surveys reach 40%. The system's payback period is less than two months, thanks to reduced costs for field crews and lab analysis.

How AI Road Inspection Works

The system processes video streams from cameras mounted on a vehicle. Each frame passes through two neural networks: a segmentation network (UNet++ for cracks) and a detection network (YOLOv8 for potholes). Results are overlaid on GPS tracks to form a defect map with a Pavement Condition Index (PCI, ASTM D6433 standard). This is 10x faster than manual inspection. The pipeline is optimized for real-time processing: p99 latency does not exceed 150 ms per frame.

What Defects Can Be Detected?

ASTM D6433 defines 20 distress types. In practice, we work with 7–8 key ones:

Why Our System Outperforms Manual Inspection?

Manual road inspection depends on the inspector's skill and lighting conditions. Our AI system uses two neural networks: UNet++ segmentation for cracks (mIoU 0.78 on test data) and YOLOv8m detection for potholes ([email protected] 0.85). PCI correlation with manual assessment is r=0.87—meaning automated approach matches expert evaluation, while being 10x faster.

Detection and Segmentation Model

import torch
import numpy as np
import segmentation_models_pytorch as smp
from ultralytics import YOLO
import cv2

class PavementInspector:
    def __init__(self, seg_model_path: str, det_model_path: str):
        # Segmentation of cracks: UNet++ with ResNet50 encoder
        # Fine-tuned on RDD2022 (Road Damage Dataset, 47k images)
        self.seg_model = smp.UnetPlusPlus(
            encoder_name='resnet50',
            encoder_weights=None,
            in_channels=3,
            classes=4,  # background, longitudinal, transverse, alligator
        )
        seg_ckpt = torch.load(seg_model_path)
        self.seg_model.load_state_dict(seg_ckpt)
        self.seg_model.eval()

        # YOLOv8m for potholes and raveling (bbox is enough)
        self.det_model = YOLO(det_model_path)

        # Segmentation class mapping
        self.seg_classes = {
            0: 'background',
            1: 'longitudinal_crack',
            2: 'transverse_crack',
            3: 'alligator_crack'
        }

        # Severity mapping for PCI-based assessment
        self.severity_thresholds = {
            'pothole': {'low': 0.01, 'medium': 0.05},    # % of frame area
            'crack': {'low': 0.02, 'medium': 0.08}
        }

    @torch.no_grad()
    def inspect(self, frame: np.ndarray) -> dict:
        h, w = frame.shape[:2]

        # 1. Crack segmentation
        input_tensor = self._preprocess(frame)
        seg_output = self.seg_model(input_tensor)
        seg_mask = seg_output.argmax(dim=1)[0].numpy()

        crack_analysis = self._analyze_cracks(seg_mask, w * h)

        # 2. Pothole detection
        det_results = self.det_model(frame, conf=0.45)
        potholes = self._analyze_potholes(det_results, w * h)

        # 3. Pavement Condition Index (simplified PCI)
        pci = self._compute_pci(crack_analysis, potholes)

        return {
            'crack_analysis': crack_analysis,
            'potholes': potholes,
            'pci_score': pci,
            'condition': self._pci_to_condition(pci),
            'seg_mask': seg_mask
        }

    def _analyze_cracks(self, mask: np.ndarray,
                          total_pixels: int) -> dict:
        analysis = {}
        for cls_id, cls_name in self.seg_classes.items():
            if cls_id == 0:
                continue
            crack_pixels = int((mask == cls_id).sum())
            ratio = crack_pixels / total_pixels
            analysis[cls_name] = {
                'pixel_count': crack_pixels,
                'area_ratio': ratio,
                'severity': 'high' if ratio > 0.08 else
                             'medium' if ratio > 0.02 else 'low'
            }
        return analysis

    def _compute_pci(self, cracks: dict, potholes: list) -> float:
        """
        PCI 0–100: 100 = perfect pavement, 0 = complete degradation.
        Simplified formula based on ASTM D6433.
        """
        deduct = 0.0
        for crack_type, data in cracks.items():
            ratio = data['area_ratio']
            if ratio > 0.08:
                deduct += 25
            elif ratio > 0.02:
                deduct += 12
            elif ratio > 0.005:
                deduct += 5

        for pothole in potholes:
            area = pothole['area_ratio']
            if area > 0.03:
                deduct += 30
            elif area > 0.01:
                deduct += 15

        return max(0, 100 - deduct)

    def _pci_to_condition(self, pci: float) -> str:
        if pci >= 85:   return 'excellent'
        elif pci >= 70: return 'good'
        elif pci >= 55: return 'fair'
        elif pci >= 40: return 'poor'
        elif pci >= 25: return 'very_poor'
        else:           return 'failed'

Mobile Inspection: Camera on a Vehicle

For public roads, we mount a camera under the front bumper or in the grille. Recording runs at 25 fps with GPS time-stamping. An accelerometer is added to automatically detect potholes via vibration.

class MobileRoadSurvey:
    def __init__(self, gps_logger, inspector: PavementInspector):
        self.gps = gps_logger
        self.inspector = inspector
        self.survey_log = []

    def process_frame_with_geotagging(self, frame: np.ndarray,
                                       timestamp: float) -> dict:
        gps_coords = self.gps.get_coords(timestamp)
        results = self.inspector.inspect(frame)

        record = {
            'timestamp': timestamp,
            'lat': gps_coords['lat'],
            'lon': gps_coords['lon'],
            'pci': results['pci_score'],
            'condition': results['condition'],
            'defects': results
        }
        self.survey_log.append(record)
        return record

Case Study: Inspection of 120 km of City Roads

Challenge: Prioritize road repairs. Our client was a city administration. Tool: Ford Transit with 4 cameras (front + 2 side + rear), GPS RTK. Over 3 days of filming, we covered 120 km.

• Frames processed: 1.2 million
• Detected: 3,400 potholes (P > 0.5), 47 km of cracks (segmentation)
• Critical sections (PCI < 25): 8.2 km—urgent repair
• Savings compared to manual survey: 12 work days → 6 hours processing + 3 hours verification, which saved over 1.8 million rubles in inspector salaries

This allowed the client to save budget and allocate resources to the most problematic sections. We apply the experience from this integration to new projects. Contact us to discuss your project.

What's Included

• Data audit: dataset collection and labeling (minimum 10,000 frames)
• Model training: fine-tuning YOLOv8 and UNet++ for your roads and climate
• GIS integration: defect coordinates mapping, PCI heatmaps
• Hardware installation: cameras, GPS, accelerometer
• Operator training and technical support
• Model warranty: free fine-tuning for seasonal changes

Process

1. Analytics: site visit, lighting assessment, reference data collection.
2. Design: model architecture selection, pipeline tuning.
3. Implementation: model training, onboard software integration.
4. Testing: trial run, results verification against manual inspection.
5. Deployment: install system on vehicle, production launch.

After deployment, the system can operate in automatic road monitoring mode, regularly sending reports to GIS.

Estimated Timelines

We guarantee quality: all models are validated on real data. Our certified engineers have implementation experience in 5 cities. Order a pilot project to see the effectiveness. Get a consultation—contact us.

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.