Video analytics for conveyor monitoring: turnkey development

With over 5 years on the market and 30+ successful projects, our team builds video analytics systems for conveyor lines turnkey— from design to deployment. The conveyor video analytics system integrates belt stop detection, product counting, and anomaly detection with high accuracy. We solve tasks that are difficult to automate with other sensors: detection of jammed objects, belt speed monitoring, detection of abnormal situations (jams, spills, product fallouts), and performance counting. One video stream replaces dozens of point sensors, covering the entire observation area. On a beverage bottling line, we reduced unaccounted product by 15%, saving the client 2.5 million rubles per year. Get a consultation on your project.

How video analytics improves conveyor monitoring accuracy?

Belt stop detection: we use optical flow Farneback — a dense method that computes motion vectors for each pixel. This allows detecting even a 10% slowdown from nominal speed. If the average motion vector in the belt region is close to zero, a stop is registered.

import cv2
import numpy as np

class ConveyorMonitor:
    def __init__(self, config: dict):
        self.belt_roi = config['belt_roi']
        self.normal_speed_range = config['belt_speed_range']
        self.alarm_callbacks = []
        self.prev_frame = None
        self.lk_params = dict(
            winSize=(21, 21),
            maxLevel=3,
            criteria=(cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 0.03)
        )

    def process_frame(self, frame: np.ndarray) -> dict:
        roi_frame = self._crop_roi(frame, self.belt_roi)
        gray = cv2.cvtColor(roi_frame, cv2.COLOR_BGR2GRAY)
        analysis = {'timestamp': get_timestamp()}
        if self.prev_frame is not None:
            flow = cv2.calcOpticalFlowFarneback(
                self.prev_frame, gray,
                None, 0.5, 3, 15, 3, 5, 1.2, 0
            )
            mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1])
            mean_speed = float(np.mean(mag))
            analysis['belt_speed'] = mean_speed
            analysis['belt_status'] = self._check_speed(mean_speed)
            analysis['obstruction'] = self._detect_obstruction(gray, flow)
        self.prev_frame = gray.copy()
        return analysis

    def _check_speed(self, speed: float) -> str:
        min_speed, max_speed = self.normal_speed_range
        if speed < min_speed * 0.1:
            return 'STOPPED'
        elif speed < min_speed * 0.5:
            return 'SLOW'
        elif speed > max_speed * 1.5:
            return 'FAST'
        return 'NORMAL'

Product counting on the conveyor

For counting, we use a cascade: YOLOv8s detector (trained on 10,000+ images of various packages) and ByteTracker. An object is counted when its center crosses a virtual line. This eliminates double counting during belt stops. Product counting accuracy of 98–99.5% is 2 times higher than ultrasonic sensors which achieve only 90–95%. Accuracy has been confirmed on 20+ lines.

class ProductCounter:
    def __init__(self, count_line_y: int, model_path: str):
        self.detector = YOLO(model_path)
        self.count_line_y = count_line_y
        self.tracker = ByteTracker()
        self.counted_ids = set()
        self.count = 0

    def process(self, frame: np.ndarray) -> int:
        detections = self.detector(frame, conf=0.5)
        tracks = self.tracker.update(detections[0])
        for track in tracks:
            cx = int((track.bbox[0] + track.bbox[2]) / 2)
            cy = int((track.bbox[1] + track.bbox[3]) / 2)
            if (track.track_id not in self.counted_ids and
                    abs(cy - self.count_line_y) < 10):
                self.count += 1
                self.counted_ids.add(track.track_id)
        return self.count

Why video analytics is more effective than sensors?

Point sensors provide information only at the installation point — they do not see jams or stops outside their zone. Video analytics covers the entire conveyor area: one 4K camera stream replaces dozens of sensors. Additionally, the system detects belt surface anomalies and spills, which are inaccessible to conventional sensors. Our experience — 5+ years in industrial video analytics, over 30 implemented projects — guarantees stable operation in workshop conditions. For example, at a cement plant, our system detected a belt blockage 2 seconds earlier than the operator, preventing 3 hours of downtime worth 150 thousand rubles.

Farneback optical flow method is described in the paper "Two-Frame Motion Estimation Based on Polynomial Expansion" (2003)

Detection of abnormal situations

For surface anomalies, we use the PatchCore model trained on reference images of conveyor belts. It detects any deviations: cracks, foreign objects, contamination. Liquid or bulk material spills are detected by a separate YOLO detector trained on synthetic data with augmentation.

class AnomalyDetector:
    def __init__(self):
        from anomalib.models import PatchCore
        self.model = PatchCore.load('conveyor_anomaly.pt')
        self.spill_detector = YOLO('spill_detector.pt')

    def check_belt_surface(self, roi: np.ndarray) -> dict:
        anomaly_result = self.model.predict(roi)
        issues = []
        if anomaly_result.pred_score > 0.7:
            issues.append({
                'type': 'surface_anomaly',
                'score': float(anomaly_result.pred_score),
                'region': self._get_anomaly_region(anomaly_result.anomaly_map)
            })
        spill_results = self.spill_detector(roi, conf=0.4)
        for box in spill_results[0].boxes:
            issues.append({
                'type': 'spill',
                'bbox': box.xyxy[0].tolist(),
                'confidence': float(box.conf)
            })
        return {'has_issues': len(issues) > 0, 'issues': issues, 'severity': 'CRITICAL' if issues else 'OK'}

Productivity heat map

Visualization of shift performance: units per hour, conveyor speed dynamics, incidents and their duration. Dashboard on Grafana with metrics from InfluxDB. We can also add belt wear prediction based on historical data.

Integration with MES/SCADA

Analytics results are transmitted to the Manufacturing Execution System via OPC-UA or REST API. Automatic line stop upon critical incidents via PLC. We support protocols OPC-UA (DA, HDA), Modbus TCP, Siemens S7. Data in JSON or Protobuf format for minimal latency.

What's included in the work

• Conveyor line audit and equipment selection
• Development of computer vision models (YOLO, optical flow, PatchCore)
• Integration with existing MES/SCADA (OPC-UA, REST API)
• Deployment on industrial server or edge device
• Operator training and documentation
• Post-release support and model retraining

How we implement the system?

Typical implementation stages:

1. Conveyor line audit and equipment selection.
2. Data collection and training of computer vision models (YOLO, optical flow, PatchCore).
3. Development and testing of detection, tracking, and anomaly algorithms.
4. Integration with MES/SCADA via OPC-UA or REST API.
5. Deployment on industrial server or edge device.
6. Operator training and handover of documentation.
7. Post-release support and model retraining.

Estimated timelines: for a single line with basic monitoring — 4–6 weeks, for 3–5 lines with anomaly detection — 8–12 weeks, for a whole workshop with MES integration — 14–20 weeks. Exact timelines are calculated after an audit. Typical payback period: 6–12 months.

Order a pilot project on one line to evaluate the effect. Get a consultation on your project — we will estimate the timeline and cost.

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.