Real-Time Object Detection: Solving the Latency Problem

Let's note: when a site has 16 surveillance cameras and the system outputs 15 FPS — that's not real-time. The pipeline described below maintains 30+ FPS on each camera under a total load of up to 32 1080p streams. We guarantee latency below 10ms through hardware decoding (NVDEC) and GPU processing. In our practice — 20+ projects in object detection. Pipeline optimization reduces GPU infrastructure costs by 2–3 times. Contact us for a free preliminary assessment of your project — it takes no more than an hour.

Why Real-Time Detection Is a Complex Technical Challenge?

The task requires a balance between accuracy, speed, and hardware load. The naive approach — running every frame through a neural network — hits the GPU limit: modern architectures (YOLOv8, RT-DETR) require 10–30 ms per inference. Without optimization, latency easily exceeds 50 ms, which is critical for robotics or security systems. The solution lies in three areas: selecting a lightweight model (YOLOv8n/m), hardware acceleration (TensorRT, NVDEC), and removing redundant load through frame skipping.

System Architecture

Camera → Frame Capture → Preprocessing → Inference → Postprocessing → Output
                ↓                              ↓
        Frame Skipping              TensorRT/ONNX Runtime
        Resize/Normalize            GPU batching

For RTSP/IP cameras, we use GStreamer or FFmpeg for stream capture with hardware decoding (NVDEC on NVIDIA):

import cv2

# Hardware-accelerated RTSP capture
cap = cv2.VideoCapture(
    'rtsp://camera_ip/stream?'
    'pipeline='
    'rtspsrc location=rtsp://camera_ip/stream !'
    'rtph264depay ! h264parse ! nvh264dec !'  # NVDEC
    'videoconvert ! appsink',
    cv2.CAP_GSTREAMER
)

Dynamic batching groups frames from multiple cameras into a single GPU pass, increasing throughput. We support batch sizes up to 32 depending on GPU memory.

How We Achieve 280+ FPS on a Single Camera?

Optimization via TensorRT is the industry standard, confirmed by the NVIDIA TensorRT Developer Guide. Converting YOLOv8 to an FP16 engine provides a 2–5x speedup over native PyTorch. We also apply dynamic batching: frames from multiple cameras are grouped into a single batch, improving GPU utilization.

from ultralytics import YOLO

model = YOLO('yolov8n.pt')
# Export to TensorRT FP16
model.export(
    format='engine',
    half=True,      # FP16 precision
    batch=1,        # or batch=4 for batching
    device=0,
    workspace=4     # GB for optimization
)

Frame skipping — we do not detect every frame. At 30 FPS video, detection on every 3rd frame (10 detections/sec) plus tracking for intermediate frames. Perceived quality is preserved.

Dynamic batching — we group frames from multiple cameras into a batch for a single GPU pass:

class MultiCameraInference:
    def __init__(self, model_path, num_cameras=8):
        self.model = load_trt_model(model_path)
        self.batch_size = num_cameras

    def process_batch(self, frames: list[np.ndarray]) -> list[list]:
        # Preprocessing batch
        batch = preprocess_batch(frames)  # [N, 3, H, W]
        # Single GPU inference for all cameras
        results = self.model.infer(batch)
        return postprocess_batch(results)

TensorRT is 3–4 times faster than native PyTorch. More details in the TensorRT documentation.

Comparison: TensorRT vs PyTorch

What Does Using TensorRT Give for Multi-Camera Systems?

For monitoring with 8–32 cameras: one A100/H100 GPU processes up to 32 1080p@30fps streams with YOLOv8n. Architecture: shared inference server (Triton) + separate capture processes for each camera. Savings on GPU infrastructure — up to 3 times compared to a naive implementation.

Throughput:

• NVIDIA T4 (16GB): 8–12 cameras 1080p with YOLOv8m
• NVIDIA A100: 24–32 cameras 1080p with YOLOv8l

How We Optimize Latency?

Pipeline latency = capture + decode + preprocess + inference + postprocess + display

Additionally, we use pipeline parallelism: capture, preprocessing, and inference execute concurrently on different GPU streams. This allows GPU utilization above 95%.

How We Implement the Solution: Step-by-Step Process

1. Requirements analysis — determine number of cameras, object classes, acceptable latency.
2. Data collection and labeling — if custom classes are needed, we prepare a dataset (1000+ frames).
3. Training and quantization — select YOLOv8n/m, train on GPU, optimize to FP16/INT8.
4. Integration with infrastructure — configure Triton Inference Server, RTSP capture, deploy in Docker.
5. Monitoring and support — Grafana dashboards, alerting, model updates.

Scope of Work and Delivery

• Architecture: capture protocol, post-processing, tracking.
• Model: YOLO selection, dataset, training, quantization to FP16/INT8.
• Inference server: configure Triton or TorchServe with batching.
• Deployment: Docker image with CUDA 12.x, Helm chart for Kubernetes.
• Documentation: API, metrics, operator instructions.
• Training: 2–4 hour workshop for your personnel.

Deployment and Monitoring

Docker container with CUDA 12.x + TensorRT. Metrics: FPS per camera, inference latency, GPU utilization, detection count per class per minute. Alerting via Prometheus + Grafana.

Cost is calculated individually based on scale and complexity. Contact us for a consultation and preliminary estimate. Savings on GPU infrastructure can reach 2–3 times.

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.