Publicly Accessible Wearable Motion Databases for Human Gait Studies

Abstract

As emerging areas of interest in gait analysis research, gait authentication and activity recognition have drawn significant attention recently, facilitated by microelectromechanical systems (MEMS) and smart devices. Machine learning models have been developed to reduce human effort and improve model accuracy in these areas. Sufficient amounts of data become critical for these applications. Unlike other fields such as image processing, in which massive data are easy to collect, human gait data is difficult to collect in large amounts, which makes publicly accessible databases in this area even more valuable. This paper aimed to summarize publicly accessible IMU-based gait databases by surveying the recent literature in human gait analysis. 199 papers were manually evaluated with sixteen data sets included (seven for gait authentication and nine for activity recognition). The detailed technical contents were described and compared in the survey to assist the audience on how to better utilize the data sets.