Human Gait Classification Model Based on Data of IMU Sensor and Multilayer Perceptron Neural Network Model

Abstract

This paper proposes a model to classify six human gaits, based on accelerometer data collected from IMU sensors that humans carry while walking, combined with Multi-Layer Perceptron Neural Network (MLPNN) model. In the MLPNN model, there are 108 inputs include 105 patterns in one gait cycle and three features of one gait cycle: skewness, entropy, the distance between high-peak and low-peak, and there are six outputs which are human gaits. To train MLPNN-model, we used the back-propagation algorithm. Six human gaits classified include: walk on the toe, walk-on heel, up the stair, downstairs, sit up, and normal-walk. The results of the proposed model are also compared with two different data mining techniques (Support Vector Machine - SVM, k-Nearest Neighbor k-NN) when classifying these six gaits. An experimental data set of the accelerometer was obtained from the IMU sensor that seven different people carried as they performed six different gaits each. To evaluation the model, we test on the Matlab-2018b software package. The results show that the accuracy of the six gaits classification is as follows: ANN-BP is 93,17%, 82.33% with k-NN, and 71.5% with SVM. This result proves that the proposed model is feasible.