CubeSat Vision-Based Localization

Overview

This repository contains the code and resources for the research project on position estimation of CubeSats using monocular vision and YOLOv8. The project explores computer vision techniques for detecting and localizing CubeSats in space using a single camera, leveraging deep learning for object detection and pose estimation.

Features

• Implementation of YOLOv8 for real-time CubeSat detection.
• Localization using a monocular camera without depth sensors.
• Camera calibration for accurate position estimation.
• Training and evaluation scripts for deep learning models.
• Experimental results and performance metrics.

Repository Structure

cubesatMonoPosEstimation/
│── data/                     # Dataset used for training and evaluation
│── models/                   # Trained YOLOv8 models
│── src/                      # Source code
│   ├── detect.py             # CubeSat detection script
│   ├── main.py               # Position estimation script MP4 video file
│   ├── pecmcv_SBC.py         # Position estimation script using USB camera
│   ├── calibraCamera.py      # Camera calibration tool
│   ├── trainYolov8n.py       # YOLOv8 training pipeline
│── results/                  # Performance results and experiment logs
│── README.md                 # Project description and instructions
│── requirements.txt          # Dependencies
│── .gitignore                # Files to be ignored in Git

Getting Started

Prerequisites

To run this project, ensure you have the following dependencies installed:

python>=3.9
torch
ultralytics
opencv-python
numpy
matplotlib

You can install them using:

pip install -r requirements.txt

Running the Detection Script

To test CubeSat detection on an image or video:

1. Go to /scripts and run download_media.ipynb cells, in sequence.
2. Run the following script:

python scripts/medirDistYoloCV2.py

Training a New Model

If you wish to train YOLOv8 on a custom dataset:

1. Go to /data and run download_dataset.ipynb cells, in sequence.
2. Run the following script:

python scripts/trainYolov8n.py

Results and Performance

Below are visualizations of the model’s performance in terms of detection accuracy and labelling:

Trainning Performance

Labelling Validation

Citation

If you use this project in your research, please cite the authors:

• vitorshaft
• igorjrd
• sergioh25

Contact

For any questions or collaborations, feel free to reach out via GitHub Issues or email: vdmrvitor@gmail.com