Fall Detection using Image Pre-Processing & Machine Learning

This project identifies whether a fall has occurred using image pre-processing, feature extraction, and machine-learning classification.
It uses traditional computer vision techniques (BRISK, BRIEF, STAR, Canny edges, histogram equalization) combined with ML models.

---

Repository Contents

1. Basic Preprocessing and Feature Extraction

File: Basic Preprocessing and feature Extraction.py

• Preprocesses images (resize, grayscale, blur, histogram equalization, edges)
• Extracts BRISK, STAR, BRIEF descriptors
• Performs K-Means clustering
• Generates:
  • Fall1.csv
  • NoFall1.csv
  • finalData.csv
  • finalFolder1.csv
  • finalFolder2.csv
  • FinalF.csv
  • Kmeans.sav

2. Classifier Training and Testing

File: Classifier and testing.py

• Loads processed feature CSV files
• Trains models:
  • Random Forest
  • Decision Tree
  • KNN
• Outputs accuracy and predictions

3. Image Background Removal Notebook

File: Image Background Remove (Preprocessing).ipynb

• Demonstrates background removal using rembg
• Helps reduce noise before feature extraction

4. Installation Instructions

File: Instructions For installing rembg.txt

• Contains steps to install the rembg library

5. Dataset Link

File: Link For Dataset Images (Fall and No Fall).txt

• Contains link to download the Fall and No-Fall dataset

---

Project Workflow

1. Image Pre-Processing

• Resize images to 100×100
• Convert to grayscale
• Apply Gaussian blur
• Perform histogram equalization
• Apply Canny edge detection
• Optional background removal

2. Feature Extraction

• Use BRISK descriptor
• Use STAR keypoint detector
• Use BRIEF descriptor
• Store extracted features in CSV files

3. Clustering (Unsupervised)

• Train K-Means with 3 clusters
• Save model as Kmeans.sav
• Convert descriptors into cluster histograms

4. Classification

• Train models using histogram data:
  • Random Forest
  • Decision Tree
  • KNN
• Compare model accuracies

---

Summary

This project provides a complete pipeline from image pre-processing to ML classification for detecting fall events using traditional computer vision and machine-learning techniques.