A new method for human activity identification using convolutional neural networks. (5th May 2023)

• Record Type: Journal Article
• Authors: Prakash, P.S.; Balakrishnan, S.; Venkatachalam, K.; Balasubramanian, Saravana Balaji
• Is Part Of: International journal of cloud computing. Volume 12:Number 2/4(2023)
• Page Start: 191
• Page End: 200
• Publication Date: 2023-05-05
• Subjects: convolutional neural network -- CNN -- people activity identification -- random forest; Cloud computing -- Periodicals; 004.678205
• Languages: English
• ISSNs: 2043-9989

Abstract

Body fitness monitoring applications are using mobile sensors to identify human activities. Human activity identification is a challenging task because of the wide availability of human activities. This paper proposes a novel technique that extracts the discriminative dimensions for human activity identification. Particularly, a novel technique with convolutional neural networks (CNN) is used for catching dependency. A deep convolutional neural network (DCNN) consists of two different types of layers, convolutional and pooling are used. The depth of each filter increases from left to right in the network. Three activities like walking, running, remaining still are collected from smart mobile sensors. The axis like x, y, and z information was transferred with column vector magnitude information and utilised for studying or training CNN. Experimental results show that CNN-based method achieves 93.67% accuracy than the baseline random forest approach's 89.20%.