sezer-muhammed/TrafficSurveillanceModelYolo

Model Card for TrafficSurveillance

This model card provides documentation for a YOLOv11m-based object detection model designed to identify pedestrians and vehicles in bird's-eye (aerial) images at 640px resolution.

Model Details

Model Description

• Developed by: Muhammed Sezer and Şevval Dikkaya
• Model type: Object Detection (YOLOv11m backbone)
• License: MIT
• Finetuned from model: Ultralytics/YOLO11m

Model Sources

• Repository: https://github.com/sezer-muhammed/Teknofest-Ulasimda-Yapay-Zeka-Veri-Seti
• Demo: https://huggingface.co/spaces/sezer-muhammed/Traffic-Object-Detection

Uses

Direct Use

This model is intended for use in surveillance, traffic monitoring, smart city applications, and public safety analysis from drone or aerial views.

Downstream Use

This model can be fine-tuned for:

• Domain-specific pedestrian tracking

Out-of-Scope Use

• Real-time ground-based pedestrian detection
• Medical, military, or privacy-invasive applications without ethical oversight

Bias, Risks, and Limitations

This model is trained on publicly sourced aerial data and may underperform in different altitudes, lighting conditions, or non-urban settings. It is also limited to detecting only two classes: pedestrian and vehicle.

Recommendations

• Apply caution in non-aerial or oblique-angle views.
• Bias due to limited diversity in dataset origin may affect generalization.
• This model should not be used in high-stakes applications without human validation.

How to Get Started with the Model

Load the model with Ultralytics framework and inference on a 640x640 aerial image.

from ultralytics import YOLO
model = YOLO('path/to/yolov11m.pt')
results = model('your_aerial_image.jpg')

Training Details

Training Data

• Dataset: Teknofest AI in Transportation Dataset
• Classes: 0 = vehicle, 1 = pedestrian
• 25,000+ images and 300,000+ labels

Training Procedure

Preprocessing

• Images resized to 640x640
• Data split into train/, test/ and dataset.yaml

Training Hyperparameters

• Mixed precision (fp16)
• Epochs: 110
• Optimizer: Adam

Summary

The model shows promise for aerial pedestrian and vehicle detection. Additional tuning recommended for deployment in non-standard aerial views.

Citation

@misc{EflatunDataset,
  author       = {Dikkaya, Şvval Belkıs and Sezer, Muhammed İzzet},
  title        = {Eflatun Takımı Teknofest Ulaşımda Yapay Zeka Yarışması Veri Seti},
  year         = {2021},
  publisher    = {GitHub},
  journal      = {GitHub repository},
  howpublished = {\url{https://github.com/sezer-muhammed/Teknofest-Ulasimda-Yapay-Zeka-Veri-Seti}},
}