Video Analytics for Traffic Monitoring: YOLO & ByteTracker

At a busy intersection, manual vehicle counting yields up to 30% error, and loop detectors break every six months. Video analytics based on YOLO and ByteTracker solves both problems: counting accuracy reaches 95–98%, and incident detection takes less than 5 seconds. By ordering such a turnkey system, you get monitoring of traffic volume by direction, flow speed, density, vehicle classification, and detection of accidents, stopped cars, and violations. Our engineers have over 10 years of experience in industrial computer vision — this guarantees stable operation in any weather. On one project for an intersection in Minsk, the system replaced 12 loop detectors, reducing maintenance costs by 60%.

How YOLO and ByteTracker ensure accurate traffic counting?

We use YOLO (You Only Look Once) — a family of neural networks for real-time object detection. Compared to classical methods (OpenCV + HOG), YOLO is 2–3 times more accurate and faster. Our base model is YOLOv8, trained on the COCO dataset and fine-tuned on recordings from city intersections. This gives classification accuracy of 92–96% and recall of 95–98%. The Ultralytics article confirms that YOLOv8 achieves mAP 0.53 on COCO — a benchmark for real-time detection. To improve robustness to weather conditions, we apply data augmentation: rain, fog, lighting changes.

Why is ByteTracker important?

To track each vehicle across frames, we implement ByteTracker. It works effectively under occlusions (when one car blocks another) and false positives. Without a tracker, correct counting and speed measurement are impossible. ByteTracker uses low-threshold detections, reducing track breaks by 30% compared to simple IoU tracking.

How do we estimate speed from video stream?

Below is the SpeedEstimator implementation, which calculates speed through position change over 5 frames. Error is ±5–10 km/h, sufficient for detecting violations and analyzing congestion.

class SpeedEstimator:
    def __init__(self, fps: float = 30.0, pixels_per_meter: float = 50.0):
        self.fps = fps
        self.ppm = pixels_per_meter
        self.track_positions = {}
        self.track_speeds = {}

    def estimate(self, track_id: int, position: tuple) -> float:
        """Speed in km/h through position change"""
        if track_id not in self.track_positions:
            self.track_positions[track_id] = []
        self.track_positions[track_id].append(position)

        history = self.track_positions[track_id]
        if len(history) < 5:
            return 0.0

        # Average displacement over last 5 frames
        recent = history[-5:]
        total_dist_px = sum(
            np.linalg.norm(np.array(recent[i]) - np.array(recent[i-1]))
            for i in range(1, len(recent))
        )
        avg_dist_px_per_frame = total_dist_px / (len(recent) - 1)
        dist_m_per_frame = avg_dist_px_per_frame / self.ppm
        speed_ms = dist_m_per_frame * self.fps
        speed_kmh = speed_ms * 3.6

        self.track_speeds[track_id] = speed_kmh
        return round(speed_kmh, 1)

Camera calibration is performed once: we measure the real distance between two points on the road and compute the pixels_per_meter coefficient. If the camera angle changes, recalibration is needed, but for stationary cameras this is a one-time procedure.

Incident Detection

The system automatically detects stopped vehicles (speed threshold <2 km/h for more than 30 seconds), hard braking, accidents, and improper pedestrian behavior. Example detector below.

class IncidentDetector:
    def __init__(self, stopped_threshold_sec: float = 30.0):
        self.stopped_vehicles = {}
        self.stopped_threshold = stopped_threshold_sec
        self.incident_cooldown = {}

    def check_incidents(self, vehicles: list[dict],
                         timestamp: float) -> list[dict]:
        incidents = []

        for vehicle in vehicles:
            tid = vehicle['track_id']
            speed = vehicle.get('speed_kmh', 0)
            pos = vehicle['center']

            if speed < 2:
                if tid not in self.stopped_vehicles:
                    self.stopped_vehicles[tid] = (pos, timestamp)
                else:
                    stopped_pos, first_seen = self.stopped_vehicles[tid]
                    duration = timestamp - first_seen
                    if duration > self.stopped_threshold:
                        if tid not in self.incident_cooldown or \
                           timestamp - self.incident_cooldown[tid] > 120:
                            incidents.append({
                                'type': 'stopped_vehicle',
                                'vehicle_id': tid,
                                'class': vehicle['class'],
                                'position': pos,
                                'duration_sec': duration
                            })
                            self.incident_cooldown[tid] = timestamp
            else:
                self.stopped_vehicles.pop(tid, None)

        return incidents

Any vehicle stopped for more than 30 seconds (configurable) is logged as an incident. This helps prevent secondary accidents and optimize tow truck dispatch. By analyzing abrupt changes in speed and position, the system can detect collisions. Additional rules can detect wrong-way driving or shoulder driving.

Vehicle Classification and Traffic Metrics

Output metrics for transportation authorities:

• PCE (Passenger Car Equivalent): truck = 2.0 PCE, bus = 1.5 PCE, motorcycle = 0.5 PCE
• LOS (Level of Service): V/C ratio → level A–F
• Speed distribution: speed histogram
• Headway: time gap between vehicles

What is included in our work?

1. Site analysis — on-site visit, assessment of camera placement, lighting, infrastructure requirements.
2. Design — model selection, processing architecture, tracker tuning, speed calibration.
3. Development and training — fine-tune YOLO for the specific site, integrate ByteTracker, create web interface.
4. Testing — run on historical recordings, measure accuracy, performance stress test.
5. Deployment and integration — install on server, connect to your software, configure alerts.
6. Documentation and training — provide instructions, train operators, 12-month warranty support.

We also version models with MLflow and monitor data drift to maintain accuracy over time. All components are licensed and compatible with traffic control systems.

Estimated Timeline

Budget savings on monitoring reach up to 50% compared to loop detectors. Get an engineer consultation and a live demo using your data. Contact us — we'll assess your project and provide a custom commercial proposal. Order a 2-week pilot project to confirm the system's effectiveness.

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.