Individual Recognition Based on Gait Using Multi-Distance Signal Level Difference Sample Entropy

Abstract

The way a person walks has unique characteristics for each individual and can be used to recognize them. There are various ways to classify characteristics for gait of each individual, one of them is the inertia sensor. The inertia sensor is used to collect data gait signals, which are angular velocity variations caused by human walking movements. Multidistance Signal Level Difference Sample Entropy is proposed in this study as a feature extraction before classifying individual gaits. MSLD is used to measure the co-occurrence of two signal samples at a distance d, and SampEn quantizes signal complexity. The MSLD Entropy produce 60 features in the form of SampEn at distances of 1 until 20 from the threeaxis. The testing procedure is carried out on the MSLD Entropy result signal for each classifier with a feature in the form of SampEn at distances of d=1-20, d=1-15, d=1-10, and d=1-5. Softmax regression as a classifier and feature at distance 1 until 20, the test results produce the greatest accuracy of 98.3%. Because a person's gait can be identified not just from one but three directions, using only one axis results in lesser accuracy than using data from all three axes.