Activity Graph Based Convolutional Neural Network for Human Activity Recognition Using Acceleration and Gyroscope Data

Human activity recognition (HAR) using smartphone sensors have been recently studied in various applications including healthcare, fitness, and smart home. Their accuracy often depends on high-quality feature design effectiveness of classification algorithms, where existing work mostly replies laborious hand-crafted shallow learning architecture. Recent deep techniques demonstrate outstanding performing automatic outperform traditional models terms accuracy. But their performance is limited by the quality volumes available labelled data. It challenging to achieve accurate multisubject HAR with only sensing This article proposes a novel optimal graph generation model incorporating framework for multiple subjects acceleration gyroscope The presents multisensory integration mechanism three-steps sorting algorithms producing graphs containing alignments neighbored signals width height. Then, we propose convolutional neural network automatically learn distinguishable features from HAR. By leveraging superior presentation correlations between human activities via graphs, learned are endowed more discriminative power. experimental evaluation was carried out several benchmark datasets (i.e., UCI, USCHAD, UTD-MHAD). results showed that our approach improved average about 5% when compared other state-of-the-art methods. Particularly towards cases (UTD-MHAD dataset 21 subjects), it achieved up 10% gain over These improvements show advantage potential method dealing complex problems