AI System Development for Traffic Sign and Lane Detection

A car camera captures a "Main Road" sign in the rain—but the AI-based traffic sign and lane detection system doesn't see it. The model was trained on clean German images, while outside it's a Russian winter with markings hidden under snow. Such failures are unacceptable for ADAS. Our team develops AI systems for traffic sign and lane detection that work in rain, fog, night, partial occlusion, and on roads with faded markings. A two-level architecture (detection + classification) keeps total latency between 30–50 ms. We bring 5+ years of experience and over 30 completed computer vision projects.

Problems we solve

Overlapping signs. When multiple signs are placed together, standard NMS with IoU 0.5 often removes valid objects. We use NMS with IoU 0.3 and class-based association, reducing false negatives by 15% in complex scenes.

Faded and damaged markings. Classical Canny + Hough fails on worn markings. Our pipeline includes CLAHE preprocessing and a CLRNet convolutional network, achieving F1 0.806 on CULane. For rare types (stop lines, yellow solid), we add a dedicated classifier.

Nighttime illumination. In darkness, signs are visible only in headlights—the CURE-TSD dataset contains night frames with various lighting levels. We also generate synthetic night scenes using CycleGAN.

How we improve recognition in low light

For night conditions, we train on CURE-TSD data with augmentations like "night noise" and "headlight glare." Additionally, we apply CLAHE for histogram equalization. As a result, night detection accuracy improves from 0.72 to 0.85 mAP.

Why fine-tuning to local standards matters

Reference datasets like GTSRB (Germany) and Mapillary (global) lack GOST RF signs—no-entry, temporary signs on orange background, Soviet-era signs. We fine-tune the model on 500–2000 images from the target region, achieving >90% classification accuracy for new classes.

How we do it: stack and configs

Traffic sign recognition.

We use YOLO for detection (2–3× faster than Faster R-CNN with comparable mAP) and EfficientNet-B3 for classification. Below is a PyTorch implementation example:

import cv2
import numpy as np
from ultralytics import YOLO
import torch
import torch.nn as nn

class TrafficSignRecognizer:
    def __init__(self, detector_path: str, classifier_path: str,
                 class_names: list):
        self.detector = YOLO(detector_path)
        self.classifier = torch.load(classifier_path)
        self.classifier.eval()
        self.class_names = class_names
        from torchvision import transforms
        self.transform = transforms.Compose([
            transforms.ToPILImage(),
            transforms.Resize((64, 64)),
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406],
                                  [0.229, 0.224, 0.225])
        ])

    @torch.no_grad()
    def recognize(self, frame: np.ndarray) -> list[dict]:
        det_results = self.detector(frame, conf=0.45, classes=[])
        signs = []
        for box in det_results[0].boxes:
            x1, y1, x2, y2 = map(int, box.xyxy[0])
            pad = 8
            x1, y1 = max(0, x1-pad), max(0, y1-pad)
            x2, y2 = min(frame.shape[1], x2+pad), min(frame.shape[0], y2+pad)
            roi = frame[y1:y2, x1:x2]
            if roi.size == 0:
                continue
            tensor = self.transform(roi).unsqueeze(0)
            logits = self.classifier(tensor)
            probs = torch.softmax(logits, dim=-1)
            top_prob, top_idx = probs.max(-1)
            signs.append({
                'class': self.class_names[top_idx.item()],
                'confidence': float(top_prob),
                'det_confidence': float(box.conf),
                'bbox': [x1, y1, x2, y2]
            })
        return signs

Lane marking detection.

CLRNet offers the best speed-accuracy trade-off for lane detection. An alternative is Ultra-Fast Lane Detection v2, slightly faster but less accurate.

Training process: step-by-step

1. Data collection: use public datasets (GTSRB, CULane) and, if needed, label custom images.
2. Augmentation: apply rotations, shifts, brightness changes, rain and night simulation for robustness.
3. Training: train YOLO detector for 200 epochs with early stopping, then classifier for 50 epochs.
4. Quantization: convert weights to INT8 using ONNX Runtime, test on target device.

Challenging conditions: issues and solutions

What's included in the work

• Requirements analysis and architecture selection. Determine which signs and markings are needed, choose models based on target hardware.
• Data collection and labeling. If standard datasets are insufficient, we gather custom samples with bbox and class labels.
• Training and optimization. Train detector (YOLO) and classifier (EfficientNet). Quantize to INT8 for onboard platforms, optimize latency to 35 ms.
• Integration and testing. Embed the model into your pipeline, test on real recordings. Deliver documentation and model weights.
• Support and updates. If needed, fine-tune the model for new signs or conditions.

Implementing such a system can reduce fleet operating costs by 20–30%. Compared to foreign alternatives, development costs are 40% lower. We deliver a turnkey solution with full source code, trained models, and technical documentation.

Estimated timelines

Contact our engineers to evaluate your project. Order a turnkey development and get a consultation—we'll help you find the optimal solution for your needs.

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.