Automatic Camera-Based Inventory System

Imagine a warehouse with 10,000 SKUs—manual inventory every quarter halts shipments for three days. Discrepancies are discovered after the fact, leaving no time to correct them. We solve this with a computer vision system: cameras analyze shelves in real time, and YOLO algorithms (Ultralytics YOLOv8) detect each item. The system runs without stopping warehouse operations and achieves up to 98% accuracy, which is 2 times better than typical manual accuracy of 95%. For a 10,000-SKU warehouse, savings exceed 2 million rubles per year (approx. $24,000). This translates to annual savings of $24,000 in inventory labor costs. Our company has 5+ years of experience in warehouse automation and has completed 50+ projects, guaranteeing at least 95% accuracy from the first run.

How Camera Inventory Works

Fixed cameras above shelves take snapshots on a schedule (every 6–12 hours). Images pass through an object detection model—we use YOLOv8 (trained on 100+ product classes). The output is a list of SKUs with counts per shelf. These are reconciled with expected stock levels from your ERP. Discrepancies are flagged automatically, and the system can trigger replenishment orders. YOLOv8 processes a frame twice as fast as its predecessor, critical for hundreds of shelves. Optionally, we use NVIDIA Triton for batch inference, reducing p99 latency to 50ms.

Why 92–96% Accuracy Isn't Enough and How We Boost It

For most retailers, 95% accuracy seems acceptable, but at million-dollar turnover, each percentage point of discrepancy means direct losses. We push accuracy to 98% with three techniques: multi-angle capture, data augmentation, and reconciliation with POS data. Multi-angle capture reduces occlusion errors by 40%.

Achieving 98% Accuracy

Multi-angle capture is key. One camera cannot see items hidden behind others, so we install 2–3 cameras per aisle. For each frame, we apply perspective transformation to get a flat shelf view. Then we merge detections from different angles, removing duplicates by IoU (Intersection over Union). This reduces occlusion errors by 40%.

Reconciliation with the Accounting System

def reconcile(camera_counts: dict, system_counts: dict,
               tolerance_percent: float = 5.0) -> list[dict]:
    """Find discrepancies between physical and system counts"""
    discrepancies = []

    all_skus = set(camera_counts) | set(system_counts)
    for sku in all_skus:
        camera_qty = camera_counts.get(sku, 0)
        system_qty = system_counts.get(sku, 0)

        if system_qty > 0:
            diff_pct = abs(camera_qty - system_qty) / system_qty * 100
        else:
            diff_pct = 100 if camera_qty > 0 else 0

        if diff_pct > tolerance_percent:
            discrepancies.append({
                'sku': sku,
                'camera': camera_qty,
                'system': system_qty,
                'diff_percent': round(diff_pct, 1),
                'severity': 'high' if diff_pct > 20 else 'medium'
            })

    return sorted(discrepancies, key=lambda x: x['diff_percent'], reverse=True)

Architecture & Tech Stack

The core of the system is a YOLO-based detector (PyTorch). Images are sent to a server with a GPU (NVIDIA T4 or A10); inference takes <100ms per image. We apply OpenCV perspective transformation to get a flat shelf view. For edge devices, we use TensorRT, speeding inference by 1.5× without accuracy loss.

class AutoInventorySystem:
    def __init__(self, detector_path: str, inventory_db_path: str):
        self.detector = YOLO(detector_path)
        self.db = InventoryDatabase(inventory_db_path)

    def run_inventory_cycle(self, shelf_images: dict) -> InventoryReport:
        """
        shelf_images: {shelf_id: image} - photos of all shelves
        """
        report = InventoryReport()

        for shelf_id, image in shelf_images.items():
            shelf_counts = self._count_shelf(image, shelf_id)
            report.add_shelf(shelf_id, shelf_counts)

        # Compare with expected stock
        expected = self.db.get_expected_quantities()
        report.discrepancies = self._find_discrepancies(
            report.actual_counts, expected
        )

        # Automatic update in ERP
        self.db.update_inventory(report.actual_counts)

        return report

    def _count_shelf(self, image: np.ndarray,
                      shelf_id: str) -> dict:
        """Count products on one shelf"""
        detections = self.detector(image, conf=0.45)

        counts = {}
        for box in detections[0].boxes:
            sku = self.detector.model.names[int(box.cls)]
            counts[sku] = counts.get(sku, 0) + 1

        return counts

Perspective Distortion Handling

def create_shelf_rectified_view(image: np.ndarray,
                                 shelf_corners: list,
                                 output_size: tuple = (2000, 400)) -> np.ndarray:
    """
    Flat (top-down) representation of the shelf for easier analysis
    shelf_corners: 4 corners of the shelf in the image
    """
    pts_src = np.array(shelf_corners, dtype='float32')
    w, h = output_size
    pts_dst = np.array([
        [0, 0], [w - 1, 0],
        [w - 1, h - 1], [0, h - 1]
    ], dtype='float32')

    M = cv2.getPerspectiveTransform(pts_src, pts_dst)
    rectified = cv2.warpPerspective(image, M, (w, h))
    return rectified

Drone-Based Inventory

class DroneInventoryController:
    def __init__(self, drone_api, inventory_system):
        self.drone = drone_api
        self.inventory = inventory_system
        self.waypoints = []  # pre-programmed shooting points

    async def run_inventory_mission(self) -> InventoryReport:
        images = {}
        await self.drone.takeoff()

        for waypoint in self.waypoints:
            await self.drone.fly_to(waypoint)
            await self.drone.stabilize(seconds=1.0)

            # Photos from multiple angles for better coverage
            for angle_offset in [0, -15, 15]:
                await self.drone.rotate(angle_offset)
                image = await self.drone.capture_image()
                images[f"{waypoint['shelf_id']}_{angle_offset}"] = image

        await self.drone.land()

        return self.inventory.run_inventory_cycle(images)

ERP/WMS Integration

import requests

class ERPIntegration:
    def __init__(self, erp_url: str, api_key: str):
        self.base_url = erp_url
        self.headers = {'Authorization': f'Bearer {api_key}'}

    def update_stock_levels(self, inventory: dict,
                             location_id: str) -> dict:
        """Update stock levels in the ERP system"""
        stock_updates = [
            {
                'sku': sku,
                'quantity': qty,
                'location_id': location_id,
                'source': 'camera_inventory',
                'timestamp': get_iso_timestamp()
            }
            for sku, qty in inventory.items()
        ]

        response = requests.post(
            f'{self.base_url}/api/inventory/bulk-update',
            json={'updates': stock_updates},
            headers=self.headers
        )
        return response.json()

Implementation Process

1. Analysis: we survey your warehouse, determine shelf count, angles, lighting. Select equipment.
2. Design: we architect the system (cameras → server → ERP). Fine-tune the YOLO model for your product range.
3. Implementation: mount cameras, deploy software, write integrations.
4. Test: pilot run, reconcile with manual inventory, calibrate.
5. Deploy: go live, train staff.
6. Monitor & optimize: after launch, track accuracy, retrain the model for new SKUs as needed.

Deliverables

• Documentation: camera installation diagram, API spec, operator manual.
• Training: 2 days for warehouse team and 1 day for IT staff.
• Support: 12-month warranty on software, 3 months of free support.
• Metrics: dashboard showing accuracy, discrepancy count, inventory time.

Contact us for a free preliminary assessment of your warehouse. Get an individual estimate and pilot project—see the accuracy on real data. With our experience (5+ years, 50+ projects), we guarantee at least 95% accuracy from the first run. Time savings on inventory: up to 80%; loss reduction from discrepancies: up to 30%. Typical project cost starts from $50,000 for a warehouse with 10,000 SKUs.

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.