Parking Monitoring System with Video Analytics

Wasting time searching for a spot and low turnover are typical pain points solved by a computer vision parking monitoring system. Compared to ultrasonic sensors that need installation per spot and often fail, video analytics uses one camera to cover up to 50 spots, with occupancy detection accuracy reaching 99%. The system runs 24/7 with IR illumination and low-light enhancement; models trained on night footage maintain up to 97% accuracy in the dark. We have 5+ years of experience and 30+ implemented projects. We guarantee stable operation and provide a free pilot on one camera to demonstrate accuracy. Annual savings on a 500-spot parking lot can reach 2 million rubles, and 100 thousand rubles on electricity thanks to using GPUs with 250W TDP. For a standard 100-space parking lot, the project cost including hardware and software starts at 500,000 rubles (approx $5,500). Contact us for a project assessment—our engineers will reach out within a day.

Problems Solved by Parking Video Analytics

Visitors spend 10–20 minutes searching for a free spot, reducing loyalty and throughput. Real-time video analytics shows available spots on each floor, guiding drivers to the nearest one. Ultrasonic and magnetic sensors fail every 2–3 years, requiring replacement and calibration. Cameras last 5–7 years, and software updates are remote. Video analytics captures license plates, entry/exit times, repeat visits—this data helps optimize pricing and plan occupancy.

Video Analytics vs Ultrasonic Sensors: Comparison

Video analytics is not only cheaper but provides more data: license plates, speed, dwell time. This enables parking optimization and revenue increase. In fact, video analytics is up to 10 times more cost-effective than ultrasonic sensors due to lower installation and maintenance costs. Annual savings on a 500-spot lot can reach 2 million rubles from reduced maintenance costs and increased turnover. Maintenance costs are cut by up to 80% compared to ultrasonic sensors. Detection speed is 2 times faster, enabling real-time updates. > Ultralytics: YOLOv8 demonstrates high detection accuracy across various scenes.

Achieving 99% Occupancy Detection Accuracy

We use computer vision algorithms based on YOLOv8 (vehicle detection) and the Ultralytics library. The model is trained on 50,000 labeled frames including night, rainy, and snowy scenes. In production, we apply post-processing: center check and bounding box overlap with the parking polygon. A time filter (vehicle stationary for over 5 seconds) eliminates false positives from passing cars. Inference runs on NVIDIA GPUs with TensorRT, with p99 latency under 150 ms per frame.

from ultralytics import YOLO
import numpy as np
import cv2
import json

class ParkingMonitor:
    def __init__(self, model_path: str, parking_config_path: str):
        self.detector = YOLO(model_path)  # vehicle detection
        self.parking_spaces = self._load_parking_config(parking_config_path)

    def _load_parking_config(self, config_path: str) -> list[dict]:
        """Load parking space configuration (polygons)"""
        with open(config_path) as f:
            config = json.load(f)

        spaces = []
        for space in config['spaces']:
            spaces.append({
                'id': space['id'],
                'polygon': np.array(space['polygon'], dtype=np.int32),
                'zone': space.get('zone', 'default')
            })
        return spaces

    def analyze(self, frame: np.ndarray) -> dict:
        """Determine occupancy of all parking spaces"""
        detections = self.detector(frame, conf=0.4,
                                    classes=[2, 5, 7])  # car, bus, truck

        # Extract vehicle centers
        vehicle_centers = []
        for box in detections[0].boxes.xyxy:
            x1, y1, x2, y2 = map(int, box)
            cx, cy = (x1 + x2) // 2, (y1 + y2) // 2
            vehicle_centers.append((cx, cy))

        # Check each parking space
        occupied_spaces = []
        free_spaces = []

        for space in self.parking_spaces:
            occupied = self._is_space_occupied(space['polygon'],
                                                vehicle_centers,
                                                detections[0].boxes.xyxy)
            space_status = {
                'id': space['id'],
                'zone': space['zone'],
                'occupied': occupied
            }

            if occupied:
                occupied_spaces.append(space_status)
            else:
                free_spaces.append(space_status)

        return {
            'total_spaces': len(self.parking_spaces),
            'occupied': len(occupied_spaces),
            'free': len(free_spaces),
            'occupancy_rate': len(occupied_spaces) / max(len(self.parking_spaces), 1),
            'spaces': occupied_spaces + free_spaces
        }

    def _is_space_occupied(self, polygon: np.ndarray,
                            vehicle_centers: list,
                            vehicle_boxes: list) -> bool:
        """Check if a vehicle is in the parking space"""
        # Method 1: vehicle center inside polygon
        for cx, cy in vehicle_centers:
            result = cv2.pointPolygonTest(polygon, (float(cx), float(cy)), False)
            if result >= 0:
                return True

        # Method 2: bounding box overlap with polygon > 40%
        space_mask = np.zeros((720, 1280), dtype=np.uint8)
        cv2.fillPoly(space_mask, [polygon], 255)
        space_area = cv2.contourArea(polygon)

        for box in vehicle_boxes:
            x1, y1, x2, y2 = map(int, box)
            vehicle_mask = np.zeros_like(space_mask)
            cv2.rectangle(vehicle_mask, (x1, y1), (x2, y2), 255, -1)

            intersection = cv2.bitwise_and(space_mask, vehicle_mask)
            overlap = intersection.sum() / 255

            if overlap / space_area > 0.4:
                return True

        return False

Labeling Parking Spaces

Initial labeling of parking space zones is done once via a web interface. Labeling a typical parking lot takes about 2 hours.

class ParkingSpaceLabeler:
    """Interactive labeling of parking spaces on a frame"""
    def __init__(self, image_path: str):
        self.image = cv2.imread(image_path)
        self.spaces = []
        self.current_polygon = []

    def interactive_label(self):
        """Start interactive labeling (for manual use)"""
        cv2.namedWindow('Label Parking Spaces')
        cv2.setMouseCallback('Label Parking Spaces', self._on_click)

        while True:
            display = self.image.copy()
            for space in self.spaces:
                cv2.polylines(display, [space['polygon']], True, (0, 255, 0), 2)
            if self.current_polygon:
                cv2.polylines(display,
                              [np.array(self.current_polygon)], False, (0, 0, 255), 2)
            cv2.imshow('Label Parking Spaces', display)

            key = cv2.waitKey(1)
            if key == ord('s') and len(self.current_polygon) >= 4:
                self.spaces.append({
                    'id': f'space_{len(self.spaces)+1}',
                    'polygon': self.current_polygon.copy()
                })
                self.current_polygon = []
            elif key == ord('q'):
                break

        return self.spaces

Integration with Navigation Systems

What capabilities does navigation integration unlock?

Our solution easily integrates into existing infrastructure:

• Digital signage: screens with a parking map, real-time updates.
• Mobile app: space booking, navigation to free spots.
• Smart traffic lights: green/red indicator above each row.
• ANPR integration: automatic entry/exit by plate, payment calculation.

Can automatic entry/exit be configured?

We provide an API for integration with any system. Data is transmitted in JSON via WebSocket or REST. For more on YOLO—the detection algorithm at the core.

Implementation Stages

1. Parking audit—on-site engineer visit, layout assessment, accuracy requirements.
2. Design—camera selection, viewing angle calculation, specification preparation.
3. Installation and configuration—camera mounting, model training on your parking lot frames.
4. Integration—connection to navigation, mobile app, access control systems.
5. Testing—accuracy verification (A/B test with real data), stress test.
6. Deployment and support—go-live, 99.9% SLA, remote monitoring.

What's Included

We deliver a turnkey project:

• Configured video analytics system with 99% accuracy.
• Source code of models and configurations.
• REST API and documentation.
• Web interface for viewing statistics and management.
• Staff training (2 hours).
• Warranty on equipment and software—12 months.

Timelines and Cost

Cost is determined individually after an audit. We offer a free pilot on one camera to demonstrate accuracy. Contact us for a project assessment—our engineers will reach out within a day. Get a consultation: reach out to us to discuss your task.

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.