Multi Attributes Recognition from Human Gait Analysis using MotionSense Dataset

Abstract

Human Gait analysis is a very prodigious and flourishing field of research nowadays, due to its immense importance in clinical and medical studies, rehabilitation, security and surveillance, crime investigation, health, sports, development of marketing applications and product optimization etc. Every human has a distinctive gait pattern, which with critical scrutiny may exhibit a lot of information about his identity and personal traits. Although researchers have made remarkable efforts in this field of research but there is a lack of work regarding sensorial gait analysis for identifying multi-attributes of a person. This paper proposes a novel framework to recognize multi-attributes i.e., user, gender, age and weight of a person based on gait analysis using smartphone built-in sensors including accelerometer, gyroscope and motion sensor. We have used an existing dataset named "MotionSense" for human activity and attributes recognition. Multi-class machine learning algorithms are applied for training the dataset. We have achieved the accuracy of 99.75% for User, 99.74% for gender, 99.61% for age and 99.74% for weight recognition respectively. Experimental results and performance evaluation of the applied machine learning classifiers reveals the efficacy of the proposed scheme.