Deep learning based indoor human activities recognition using channel state information

Khoo, Bee Sze (2020) Deep learning based indoor human activities recognition using channel state information. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Deep learning based indoor human activities recognition using channel state information.pdf - Submitted Version Download (3MB)

Abstract

Do you ever believe that the human activities can be recognized by only using Wi-Fi signals? In this project, a Wi-Fi based activity recognition using deep neural networks is proposed to recognize the indoor human activities. Compared to the traditional human activities recognition approaches, which employed the used of the camera and the sensors, the proposed method is unobstructive, respect to the individua l‘s privacy and works without affected by the lighting condition. Moreover, the complexity of feature extraction processing is simplified due to the powerful inference of deep neural networks. In this project, the standard Unidirectional Long Short Term Memory (Uni-LSTM) had been developed for human activity recognition. Besides, more complex architectures such as Bidirectional LSTM (Bi-LSTM) and cascaded (Cas-LSTM) has also been investigated. An accuracy of 98.33% had been achieved by using self-collected dataset using the proposed Bi-LSTM model. All of the experiment results including the effects of human activities on the Wi-Fi signal and the performance of the proposed networks have been analyzed and evaluated in the project thesis.

Actions (login required)

View Item