Feature Extraction, Classification and Prediction for Hand Hygiene Gestures with KNN Algorithm

TL;DR

This study uses computer vision to extract hand features from videos of hand-washing gestures and applies a KNN classifier, achieving over 95% accuracy in recognizing six WHO-recommended gestures.

Contribution

It introduces a novel feature extraction approach for hand gestures and demonstrates the effectiveness of KNN in classifying hand hygiene gestures with high accuracy.

Findings

Achieved >95% classification accuracy

Identified optimal K=3 for KNN

Validated method with cross-validation

Abstract

There are six, well-structured hand gestures for washing hands as provided by World Health Organisation guidelines. In this paper, hand features such as contours of the hands, the centroid of the hands, and extreme hand points along the largest contour are extracted for specific hand-washing gestures with the use of a computer vision library, OpenCV. For this project, a robust dataset of hand hygiene video recordings is built with the help of 30 research participants. In this work, a subset of the dataset was used as a pilot study to demonstrate the effectiveness of the KNN algorithm. Extracted hand features saved in a CSV file are passed to a KNN model with a cross-fold validation technique for the classification and prediction of the unlabelled data. A mean accuracy score of >95% is achieved and proves that the KNN algorithm with an appropriate input value of K=3 is efficient for hand hygiene gestures classification.