A custom object detection system designed to identify traffic signs within Silpakorn University’s Sanam Chandra Campus — where many signs differ significantly from standard traffic signage, making public datasets unsuitable. This project builds a full end-to-end pipeline from dataset creation to model training and evaluation using YOLOv8.
This project focuses on detecting campus-specific traffic signs using computer vision. Due to the unique nature of the signs found within the university, the model relies entirely on a self-built dataset, tailored to the environment of Silpakorn University's Sanam Chandra Campus.
Below are links to the dataset and model results from this project:
-
📦 Dataset on Roboflow Universe (Created by Me):
-
📁 Model Results on Google Drive (Trained Models from This Project):
🔗 View My Model Outputs
- Captured on-campus video footage containing traffic signs.
- Extracted individual image frames from video files.
- Annotated images using Roboflow, labeling various types of signs.
- Performed image resizing and augmentation to enhance dataset quality.
- Exported the dataset in YOLOv8-compatible format.
Trained five variants of the YOLOv8 model to compare performance:
yolov8nyolov8syolov8myolov8lyolov8x
Each model was trained using the prepared dataset with consistent preprocessing settings for fair comparison.
Model experiments were tracked using Comet ML, allowing for:
- Real-time training monitoring
- Metrics comparison across models
- Visualization of model performance
- YOLOv8 | Object detection model
- Roboflow | Annotation, augmentation, export
- Comet ML | Training monitoring and analytics
- OpenCV | Frame extraction from video