Privacy-Preserving Abnormal Gait Detection Using Computer Vision and Machine Learning

Abstract

Gait analysis plays a pivotal role in diagnosing a spectrum of neurological and musculoskeletal disorders. Variations in gait patterns often serve as early indicators of underlying health conditions, underscoring the importance of precise and timely analysis for effective intervention and treatment. In recent years, computer vision techniques have emerged as robust tools for automated gait analysis, offering non-invasive, cost-effective, and scalable solutions. However, existing approaches often overlook the critical aspect of privacy preservation. In this study, we propose the world’s pioneering computer vision-based abnormal gait detection system with a privacy-preserving mechanism. Specifically, we extract 2D skeletons from encrypted images using a deep neural network model, which is facilitated by an optical system incorporating a custom-made refractive optical element. These extracted features are then fed into machine learning models for the detection of normal versus abnormal gait patterns. Evaluations across various models including random forest, decision tree, k-nearest neighbor, support vector machine, neural network, and convolutional neural network reveal that the random forest model attains the highest classification performance based on 2D skeletons extracted from encrypted images.