Machine Learning Project Documentation: YOLOv8 for Gadget Detection

1. Title Page

Title: YOLOv8 for Gadget Detection in Oloid Presence Detection Feature (Windows Login Application) Author(s): [Your Name(s)] Date: [DD/MM/YYYY]

2. Abstract

This document outlines the implementation of the YOLOv8 machine learning model for gadget detection within the Oloid Presence Detection feature of the Windows Login application. It covers the problem statement, methodology, model selection, training, and evaluation.

3. Introduction

3.1 Problem Statement

The goal is to enable real-time gadget detection to enhance security in the Windows Login application by detecting unauthorized devices in the user’s vicinity.

3.2 Objective

Develop a robust and efficient object detection model using YOLOv8 to identify gadgets such as mobile phones, tablets, and smartwatches during user authentication.

3.3 Motivation

Unauthorized devices near a system pose a security risk. By integrating gadget detection into the Windows Login process, Oloid enhances security by ensuring a trusted computing environment.

3.4 Scope

The implementation is focused on real-time detection and classification of gadgets using a pre-trained YOLOv8 model, fine-tuned on relevant datasets.

4. Data Collection & Preprocessing

4.1 Data Source

Custom dataset collected from real-world scenarios, including images of different gadgets in various environments.
Augmented with publicly available datasets for electronic device detection.

4.2 Exploratory Data Analysis (EDA)

Dataset distribution across different gadget categories.
Data augmentation techniques applied to improve model robustness.

4.3 Feature Engineering

Bounding box annotations for gadgets.
Normalization and resizing of images for YOLOv8 input compatibility.

4.4 Data Splitting

Training Set: 70%
Validation Set: 20%
Test Set: 10%

5. Model Selection

5.1 Algorithm Considered

YOLOv8 (You Only Look Once - Version 8) selected for real-time object detection.

5.2 Justification

High-speed performance and accuracy suitable for real-time detection.
Optimized for edge deployment with minimal computational overhead.

5.3 Hyperparameter Tuning

Batch Size: [Value]
Learning Rate: [Value]
Number of Epochs: [Value]
Anchor Box Adjustments: Fine-tuned for gadget detection.

6. Implementation Details

6.1 Tools & Frameworks

Python
Ultralytics YOLOv8
OpenCV
TensorFlow/PyTorch
Windows SDK (for integration with Windows Login application)

6.2 Code Snippets

from ultralytics import YOLO  # Load the pre-trained YOLOv8 model model = YOLO('yolov8n.pt')  # Run inference on an image results = model('test_image.jpg')  # Display results results.show()

6.3 Model Training

Trained on GPU-enabled hardware for accelerated learning.
Augmented dataset for robustness against different lighting conditions and device orientations.

7. Evaluation & Results

7.1 Performance Metrics

Mean Average Precision (mAP50): [Value]
Precision: [Value]
Recall: [Value]
F1-Score: [Value]

7.2 Comparison with Baseline

Compared against previous gadget detection methods (if applicable).

7.3 Visualizations

Confusion Matrix
Precision-Recall Curve
Sample Detection Outputs

8. Challenges & Limitations

8.1 Problems Faced

Difficulty in detecting smaller devices in cluttered environments.
Variability in gadget designs affecting detection accuracy.

8.2 Limitations

Model performance is dependent on dataset diversity.
Potential false positives in noisy environments.

9. Conclusion & Future Work

9.1 Summary

YOLOv8 successfully enables real-time gadget detection for Oloid’s Windows Login application, enhancing security by preventing unauthorized devices from being present during authentication.

9.2 Future Improvements

Further model optimization for improved detection accuracy.
Expansion of dataset to include more gadget variations.
Integration of additional security measures based on detected gadgets.

10. References

Ultralytics YOLOv8 Documentation
Research papers on real-time object detection
Public datasets used for training