SafeSpace

Overview

SafeSpace is an iOS application that uses advanced machine learning to detect objects in space station environments. Leveraging the YOLOv8 object detection model, the app helps astronauts and space personnel identify and categorize objects in various lighting conditions and environments that might be encountered aboard a space station.

The current implementation is trained on the Falcon synthetic dataset and can detect three critical safety-related object classes:

• Fire extinguishers
• Toolboxes
• Oxygen tanks

Features

Real-time Object Detection

• Camera-based detection with advanced YOLOv8 model
• Real-time processing and feedback
• Detection metrics including accuracy and processing time

Simulation Laboratory

• Test object detection in simulated space environments
• Adjustable lighting and occlusion conditions
• Multiple environment presets (Normal Station, Dim Lighting, Emergency Lighting, etc.)
• Interactive zoom functionality for detailed analysis

Detailed Analysis

• Comprehensive detection results with confidence scores
• Performance metrics for detection accuracy
• List of detected objects with classification
• Classification of fire extinguishers, toolboxes, and oxygen tanks with confidence scores

Technical Details

Machine Learning

• Implements YOLOv8 (You Only Look Once) object detection model
• Trained on the Falcon synthetic dataset of space station environments
• Optimized for iOS using Core ML framework
• Fast inference time (typically under 50ms)
• Specialized in detecting safety equipment in space habitats

Model Performance

The optimized YOLOv8m model was trained on the HackByte_DataSet (Falcon synthetic dataset) specifically for space station environments. Below are the performance metrics:

Confusion Matrix

The confusion matrix shows the model's classification accuracy across the three object classes:

Precision-Recall Curve

The PR curve demonstrates the balance between precision and recall across different confidence thresholds:

Training Results

The results graph shows the model's performance metrics during training, including mAP (mean Average Precision), precision, and recall:

Key performance indicators:

• mAP@50: 0.812(all classes)
• Precision: 1.00 at 0.952(all classes)
• Recall: 0.85 at 0.000(all classes)
• F1-Score: 0.810 at 0.344(all classes)

iOS Technologies Used

• SwiftUI for modern, responsive UI
• AVFoundation for camera handling
• Core ML for on-device machine learning
• Combine for reactive state management

Requirements

• iOS 16.0 or later
• iPhone or iPad with camera
• Xcode 14+ (for development)

Installation

1. Clone the repository

   git clone https://github.com/yourusername/SafeSpace.git
   

2. Install dependencies.

3. Open the project in Xcode

   cd SafeSpace
   open SafeSpace.xcodeproj
   

4. Build and run on a physical device for full functionality

Usage

Detection Mode

1. Launch the app and navigate to the Detection tab
2. Grant camera permissions when prompted
3. Point the camera at objects to detect
4. Tap the capture button to analyze the current frame
5. Review detected objects and their confidence scores

Simulation Lab

1. Navigate to the Simulation tab
2. Upload an image using the photo picker or use a previously captured image
3. Select an environment preset or adjust lighting/occlusion levels manually
4. Press "Start Simulation" to run the detection model
5. Pinch to zoom in/out on the results for detailed inspection
6. Double-tap to reset zoom level

License

[MIT]

Acknowledgments

• YOLOv8 developed by Ultralytics
• HackByte_DataSet (Falcon synthetic dataset) for space station environments

Screenshots

Camera Detection Interface (left) and Simulation Lab (right)

Detection Results (left) and Object Detection Flow (right)

Simulation Workflow (left) and Additional View (right)