Design and development of an improved Wi-Fi based human activity recognition using LSTM

Chan, Jacky Chen Hsien (2021) Design and development of an improved Wi-Fi based human activity recognition using LSTM. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Design and development of an improved Wi-Fi based human activity recognition using LSTM.pdf - Submitted Version Download (3MB)

Abstract

Wi-Fi-based human motion sensor technology has been broadly developed from time to time. By comparing with traditional human behaviour recognition system, the benefits of using Wi-Fi-based motion technology are that it had large coverage of signal with unconstraint by any obstacles in terms of dead angle and sensitivity of light. This paper proposes an improved Wi-Fi based human activity recognition by using a deep learning method of Long Short-Term Memory (LSTM). A total of 139 data from first datasets and 140 from second datasets were obtained from internet have been tested by LSTM. Before testing, denoise process and feature extraction by discrete wavelet transform (DWT) were underwent first to extract significant feature of dataset. The performance of the extracted features will be analysed by using Bi-LSTM. The result showed 96.4% and 89.3% of top highest accuracy rate by using proposed signal pre-processing method. For the first and second raw dataset, they achieved 64.3% and 60.7% respectively since the raw datasets that had not going through process of denoise and feature extraction. Besides, the paper also discussed performance of consistency and time processing of LSTM classification between the datasets. Some recommendation and limitation also provided before the end of the paper.

Actions (login required)

View Item