[Paper Review] Feature Detection for Hand Hygiene Stages

Rashmi Bakshi, Jane Courtney+2 more|arXiv (Cornell University)|Aug 6, 2021

Hand Gesture Recognition Systems8 references4 citations

TL;DR

This paper proposes a computer vision pipeline for detecting hand hygiene stages using feature detection algorithms on a custom-built aluminum rig dataset. It evaluates Harris, Shi-Tomasi, and SIFT detectors on static images of the 'rubbing palms' stage, demonstrating feasibility for future video-based deep learning classification of WHO-recommended handwashing steps.

ABSTRACT

The process of hand washing involves complex hand movements. There are six principal sequential steps for washing hands as per the World Health Organisation (WHO) guidelines. In this work, a detailed description of an aluminium rig construction for creating a robust hand-washing dataset is discussed. The preliminary results with the help of image processing and computer vision algorithms for hand pose extraction and feature detection such as Harris detector, Shi-Tomasi and SIFT are demonstrated. The hand hygiene pose- Rub hands palm to palm was captured as an input image for running all the experiments. The future work will focus upon processing the video recordings of hand movements captured and applying deep-learning solutions for the classification of hand-hygiene stages.

Motivation & Objective

• To develop a robust, reproducible dataset for hand hygiene stage recognition using a custom aluminum rig.
• To evaluate traditional computer vision feature detectors (Harris, Shi-Tomasi, SIFT) on static images of hand hygiene poses.
• To establish a foundation for future video-based classification of WHO handwashing stages using deep learning.
• To enable precise hand pose extraction and feature localization for complex hand movement analysis.
• To support infection control by enabling automated monitoring of hand hygiene compliance.

Proposed method

• Constructed a calibrated aluminum rig to standardize hand positioning and lighting for consistent image capture.
• Collected a dataset of static images depicting the 'rub hands palm to palm' stage of handwashing.
• Applied Harris corner detector, Shi-Tomasi corner detector, and SIFT feature detection algorithms to identify key hand features.
• Used image processing techniques to extract and localize salient features from hand regions in the input images.
• Evaluated detection performance on a single, well-defined hand hygiene pose to assess algorithmic feasibility.
• Laid the groundwork for extending the approach to video sequences using deep learning models.

Experimental results

Research questions

• RQ1 How effective are traditional feature detection algorithms (Harris, Shi-Tomasi, SIFT) in identifying key hand features during a specific hand hygiene stage?
• RQ2 Can a controlled, physical rig setup produce a reliable and repeatable dataset for hand hygiene stage recognition?
• RQ3 To what extent can static image analysis of hand poses support the development of video-based classification systems?
• RQ4 What are the limitations of classical computer vision methods in capturing the dynamic nature of handwashing motions?
• RQ5 How can feature detection be leveraged to enable automated monitoring of WHO-recommended hand hygiene steps?

Key findings

• The aluminum rig enabled consistent, repeatable image capture under controlled conditions, supporting reliable dataset creation.
• Harris, Shi-Tomasi, and SIFT detectors successfully identified salient features in the 'rub hands palm to palm' hand pose.
• Preliminary results demonstrate the viability of classical feature detection for hand hygiene stage analysis.
• The study confirms that static image analysis using these algorithms can serve as a strong baseline for future video-based deep learning approaches.
• The dataset and methodology provide a foundation for classifying the full sequence of WHO-recommended handwashing steps.
• Future work will extend the approach to video sequences and leverage deep learning for improved classification accuracy.