HawkWatch

Drone YOLO Object Detection 📦🛩️ This repository contains a full pipeline for training a YOLOv8 model on custom drone footage data for object detection and line-crossing analysis. The implementation uses the Ultralytics YOLOv8 framework and is tailored for lightweight and fast inference using the YOLOv8 Nano variant.

🧠 Project Overview This project trains a YOLOv8 model to detect objects from drone footage. The detections can later be used for advanced tasks like object counting, trajectory tracking, and boundary-line crossing analytics.

📁 Directory Structure The dataset is organized as per the YOLO format. Below is the visual layout:

drone_yolo_dataset/ ├── train/ │ ├── images/ → Training images │ └── labels/ → Corresponding YOLO labels ├── val/ │ ├── images/ → Validation images │ └── labels/ → Corresponding YOLO labels ├── test/ │ ├── images/ → Test images │ └── labels/ → Corresponding YOLO labels data.yaml → Configuration file linking datasets and class names 📄 data.yaml Configuration This file defines the dataset paths and class names for the model:

yaml train: drone_yolo_dataset/train/images val: drone_yolo_dataset/val/images test: drone_yolo_dataset/test/images

nc: 1 # Number of classes names: ['drone_object'] # Class names Make sure this file is in the project root and correctly references the dataset folder paths.

🚀 Training Script A minimal script to train the model using Ultralytics:

from ultralytics import YOLO import os

Create dataset structure if not present

required_dirs = [ "drone_yolo_dataset/train/images", "drone_yolo_dataset/train/labels", "drone_yolo_dataset/val/images", "drone_yolo_dataset/val/labels", "drone_yolo_dataset/test/images", "drone_yolo_dataset/test/labels" ] for d in required_dirs: os.makedirs(d, exist_ok=True)

Load YOLOv8 Nano model

model = YOLO("yolov8n.yaml")

Train the model

model.train( data="data.yaml", epochs=50, imgsz=640, batch=16, name="drone_yolo_model", project="runs/train", workers=4, val=True )

Evaluate the model

metrics = model.val(data="data.yaml", split="test") print("Test metrics:", metrics)

Run predictions

results = model.predict(source="drone_yolo_dataset/test/images", save=True) ✅ Requirements

pip install ultralytics Ensure that torch and opencv-python are also installed for smooth training and prediction.

📊 Output Trained model weights: runs/train/drone_yolo_model/weights/best.pt

Prediction results saved in the current directory (with bounding boxes)

📌 Notes Label format follows YOLO: [class_id x_center y_center width height], all normalized.

Ensure correct image-label pairing (same filenames, different extensions).

You can visualize predictions using model.predict(...) or explore results using the YOLO UI.