Efficient Frequency Domain Feature Extraction Model using EPS and LDA for Human Activity Recognition

Abstract

Activity identification based on machine learning for human computing aims to understand or capture the state of human behavior, its environment, and integrate user by exploiting distinct types of sensors to give adjustment to the exogenous computing system. The ascent of universal computing systems requires our environment a solid requirement for novel methodologies of Human Computer Interaction (HCI). The recognition of human activities, commonly known as HAR can play a vital task in this regard. HAR has an appealing use in the health-care system and monitoring of Daily Living Activities (DLA) of elderly people by offering the input for the development of more interactive and cognitive environments. This paper is presenting a model for the recognition of Human Activities. In this proposed model, the Enveloped Power Spectrum (EPS) is used for extracting impulse components of the signal, and the Linear Discriminant Analysis (LDA) is used as a dimensionality reduction procedure to extract the discriminant features for human daily activity recognition. After completing EPS feature extraction techniques, LDA is performed on those extracted spectra for extracting features using the dimension reduction technique. Finally, the discriminant vocabulary vector is trained by the Multiclass Support Vector Machine (MCSVM) to classify human activities. For validating the proposed scheme, UCI-HAR datasets have been implemented which demonstrates higher recognition accuracy which has been acknowledged.