Human Activity Recognition Using Smartphone Sensor Data Based on Hybrid Model

Abstract

Accelerometers, gyroscopes, GPS, and various sensors have become widespread in smartphones. In accordance with this trend, many studies are actively conducting research on detecting and recognizing human activities using data acquired from smartphone sensors without separate attachments. Human activity recognition technology is gaining attention not only in specific fields such as security facilities and hospitals but also in everyday life and entertainment. In previous studies, researchers manually extracted effective features for activity recognition from raw signals acquired by sensors or utilized artificial neural networks to automatically extract features. However, no method showed significantly superior recognition performance compared to others. In this study, a hybrid CNN model that uses both handcrafted features and automatically extracted features using CNN is proposed. Experimental results on the UCI-HAR dataset representing six types of activities showed an impressive accuracy of 97.33%. It shows that the proposed approach is effective in recognizing human activity.