Sleep Position Classifier

A machine learning-based Sleep Position Classifier designed to identify one of three sleep postures: Left, Supine, and Right. The system leverages classical machine learning (SVM, KNN) and deep learning (CNN) models, using both handcrafted HOG features and raw image inputs. Additionally, a novel Neutrosophic conversion step is applied for enhanced preprocessing using Falsity, Indeterminacy, and Truth maps.

---

Project Overview

This project aims to classify sleep positions based on image data. The complete pipeline includes image preprocessing using Neutrosophy, feature extraction using HOG (Histogram of Oriented Gradients), and classification using:

• Support Vector Machine (SVM)
• K-Nearest Neighbors (KNN)
• Convolutional Neural Network (CNN)

---

📁 Dataset

• Images of individuals in three sleep positions: Left, Supine, and Right
• Stored in structured class-wise directories
• Images are initially processed for Neutrosophic mapping to reduce uncertainty

---

🧠 Preprocessing

Neutrosophic Image Conversion

All input images undergo transformation using the Neutrosophic theory, resulting in three grayscale components:

• Falsity
• Indeterminacy
• Truth

These components are used in downstream model training and evaluation.

HOG Feature Extraction

• Images are transformed into HOG features with 1764 feature dimensions
• Used as input for traditional ML models like SVM and KNN

---

⚙️ Models and Methods

Model	Input Type	Description	Accuracy
SVM	HOG Features (1764D)	Linear classification on handcrafted features	96%
KNN	HOG Features (1764D)	K=7 neighbors for classification	95.67%
CNN	Raw Images	End-to-end deep learning with ConvNet architecture	96%

---

🛠️ Tech Stack

• Python 3.10+
• OpenCV - Image processing
• scikit-image - HOG feature extraction
• scikit-learn - SVM, KNN models
• TensorFlow / Keras - CNN model building
• NumPy, Matplotlib - Data handling and visualization
• Custom Neutrosophic Module - Preprocessing logic
• HOG - Algorithm

---

🚀 Getting Started

1. Clone the repository

   git clone https://github.com/karthiksuki/Sleep-Position-Classification
   cd sleep-position-classifier

2. Install dependencies

   pip install -r requirements.txt

3. Run Neutrosophic Preprocessing

   python neutroscophy_conversion/neutroscopic_conversion.py

4. Train Models

   python models/model_svm.py
   python models/model_knn.py
   python models/model_cnn.py

5. Make Predictions

   python Prediction_Model.py

---

📊 Results

• SVM: 96% Accuracy
• KNN: 95.7% Accuracy
• CNN: 96% Accuracy

---

🧪 Future Improvements

• Integration with real-time video-based prediction
• Expand dataset with various body types and sleepwear
• Include edge-case postures (e.g., fetal, prone)

---

👨‍💻 Author

• Dr. Nagarajan Deivanayagampillai, Post-Doctorate, Ph.D
• Karthikeyan Ganesh - Linkedln