Practical experience building labeled datasets from scratch for multi-class vehicle detection using YOLO11
Practical experience building labeled datasets from scratch for multi-class vehicle detection using YOLO11. Starting from raw CCTV footage, frames were extracted, consistent labeling guidelines defined, and bounding boxes annotated for four vehicle types: car, motorcycle, truck, and bus. Utilized Roboflow to produce YOLO-format label files and managed the full dataset pipeline including train/val/test splits and data.yaml configuration. Trained and evaluated the model end-to-end, applying data augmentation strategies such as flipping, cropping, and brightness adjustment to handle varying lighting and traffic conditions. The resulting detection system was applied in a project analyzing the relationship between vehicle counts and air pollution levels, alongside an LSTM forecasting model developed for time-series prediction.