Development of License Plate Recognition System (ANPR/LPR)

Imagine a camera at a shopping mall parking lot capturing an entering vehicle. The system must recognize the plate in milliseconds and decide whether to raise the barrier or add it to a blacklist. If OCR fails, you get a traffic jam at the entrance and negative visitor experience. How do you build an ANPR/LPR system that reliably works in rain, at night, and at speeds up to 60 km/h? We have refined this pipeline over years and share our proven architecture. Our system saves up to 30% budget through open-source models and inference optimization.

How the Two-Stage License Plate Recognition Pipeline Works

Video/Photo → Vehicle Detection → License Plate Detection → OCR → Database

The two-stage approach (vehicle → plate) is more accurate than one-stage because it handles different plate formats from different countries. The first stage uses YOLO to detect vehicles, the second uses a specialized model to detect the plate within the crop.

from ultralytics import YOLO
from paddleocr import PaddleOCR
import cv2
import numpy as np
import re

class ANPRSystem:
    def __init__(self,
                 vehicle_model: str = 'yolov8l.pt',
                 plate_model: str = 'plate_detector.pt'):
        self.vehicle_detector = YOLO(vehicle_model)
        self.plate_detector = YOLO(plate_model)  # fine-tuned on license plates
        self.ocr = PaddleOCR(
            use_angle_cls=True,
            lang='en',
            rec_algorithm='SVTR_LCNet'
        )

    def process(self, frame: np.ndarray) -> list[dict]:
        # Vehicle detection
        vehicles = self.vehicle_detector(frame, classes=[2, 3, 5, 7],  # car/moto/bus/truck
                                          conf=0.5)
        results = []

        for vehicle_box in vehicles[0].boxes.xyxy:
            x1, y1, x2, y2 = map(int, vehicle_box)
            vehicle_crop = frame[y1:y2, x1:x2]

            # License plate detection in vehicle crop
            plates = self.plate_detector(vehicle_crop, conf=0.5)

            for plate_box in plates[0].boxes.xyxy:
                px1, py1, px2, py2 = map(int, plate_box)
                plate_crop = vehicle_crop[py1:py2, px1:px2]

                # OCR for plate
                plate_text = self._recognize_plate(plate_crop)

                if plate_text:
                    results.append({
                        'plate': plate_text,
                        'vehicle_bbox': [x1, y1, x2, y2],
                        'plate_bbox': [x1+px1, y1+py1, x1+px2, y1+py2],
                        'confidence': float(plates[0].boxes.conf[0])
                    })

        return results

    def _recognize_plate(self, plate_img: np.ndarray) -> str | None:
        # Preprocessing
        plate_img = self._preprocess_plate(plate_img)

        result = self.ocr.ocr(plate_img, cls=False)
        if not result or not result[0]:
            return None

        text = ''.join([line[1][0] for line in result[0]])
        text = re.sub(r'[^A-Z0-9А-Я]', '', text.upper())

        # Validation of Russian plate format
        if re.match(r'^[АВЕКМНОРСТУХ]\d{3}[АВЕКМНОРСТУХ]{2}\d{2,3}$', text):
            return text

        return text if len(text) >= 6 else None

Why Image Preprocessing Matters for OCR Quality

OCR accuracy directly depends on how well the plate crop is prepared. We use scaling to a height of 64 pixels, angle alignment, and brightness normalization. This reduces error rate by 15–20% compared to raw frames.

def _preprocess_plate(self, image: np.ndarray) -> np.ndarray:
    # Scale to standard height
    target_h = 64
    scale = target_h / image.shape[0]
    new_w = int(image.shape[1] * scale)
    image = cv2.resize(image, (new_w, target_h), interpolation=cv2.INTER_CUBIC)

    # Convert to grayscale
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

    # Brightness normalization
    normalized = cv2.normalize(gray, None, 0, 255, cv2.NORM_MINMAX)

    return normalized

How We Handle Different Plate Formats

Russian plates: X000XX00[0] (standard), X000XX000 (transit). Additional formats: customs, diplomatic, military. For international systems, we use multilingual OCR and multiple validation regex patterns. We have accumulated a library of over 20 masks for CIS and European countries.

Real-World Case: Shopping Mall Parking Lot with 8 Cameras

For a large shopping center, the system needed to handle a flow of 30 cars per minute, operate 24/7, and integrate with existing barriers. We deployed the two-stage pipeline on a server with GPU T4. The recognition accuracy reached 98%, false positives under 1%. Response time was 45 ms per frame. After a year of operation, the system required no retraining — only periodic camera calibration.

Comparison: Our System vs Typical OpenALPR Solutions

Our pipeline is twice as fast at the same accuracy: 45 ms vs 95 ms on T4. Through fine-tuning YOLO and PaddleOCR, we achieve 98% accuracy compared to 93% for OpenALPR on challenging plates. Moreover, we support more formats — over 20 masks vs 5 standard.

Step-by-Step ANPR/LPR System Deployment

1. Audit of installation site and camera selection (resolution, IR illumination).
2. Dataset collection: 5,000+ frames in various conditions for model fine-tuning.
3. Training detection and OCR models on a compute cluster (typically 2–3 days on GPU A100).
4. Integration with access control via REST API, Redis setup for LPR lists.
5. One-week testing with real traffic, threshold adjustments.
6. Deployment on the client's server, documentation, and staff training.

Production Performance

What's Included in Turnkey Development

• Site analysis and camera selection
• Training/fine-tuning of detection and OCR models
• Preprocessing and postprocessing configuration
• REST API for integration with access control and databases
• Redis for hot lists (whitelist/blacklist)
• PostgreSQL with pg_trgm for fuzzy search (accounts for OCR errors: 0/O, I/1, B/8)
• Documentation and staff training
• 6-month warranty support

Implementation Timelines

The cost is calculated individually — depends on the number of cameras, required accuracy, and depth of integration. Our engineers hold MLOps certifications and have a combined 10+ years of experience in Computer Vision. Contact us for a free project assessment — we evaluate your project within one business day. We guarantee transparent results and adherence to deadlines.

Technologies used: YOLOv8, PaddleOCR, PyTorch, Redis, PostgreSQL.

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.