Behavior pattern recognition based on convolutional neural network using accelerometer and gyroscope sensors
JunHyuk Kang, Yeseul Lim, Ahyoung Choi*
Department of Software, Gachon University, Seongnam, Republic of Korea
Abstract: Recently, research on recognizing a human behavior pattern based on accelerometer and gyroscope signals has been done. However, the existing studies were mainly conducted to analyze the extracted features for behavior pattern recognition as inputs. In this study, we propose a model that recognizes and classifies behavior patterns by using raw signals including noise as inputs of the convolutional neural network(CNN) model to which optimal hyperparameters are applied through the grid search algorithm. The behavior patterns were divided into 4-classes. To evaluate the proposed model, 12 data were collected, and it was confirmed that the average accuracy was 93%.
Keywords: wearable devices; human behavior; accelerometer signal; gyroscope signal; convolutional neural network;