A Deep Bidirectional LSTM Recurrent Neural Networks For Identifying Humans Indoors Using Channel State Information

Human identification is extremely crucial in the field of human-computer interaction. A number of studies on human identification have been done which includes using face and human gaits. Human beings have unique body structures and gaits patterns, so they induce different signal propagation paths which results in producing unique CSI signatures. These unique CSI signatures could be mapped with each individual person portraying them and by thus uniquely identifying a person. Since there are limited empirical research conducted on human identification using Wi-Fi and deep learning models, we propose a Deep bidirectional LSTM recurrent Neural networks (DBD-LSTM-RNN) for Identifying humans indoors using channel state information. A deep bidirectional LSTM-RNN model that segments the signals to determine the start and the end of human gait and map them with the appropriate body structure is deployed. Furthermore, we employed the Chebyshev filter to reduce noise on the collected CSI data. Lastly, the model is tested and evaluated using the data we have collected. The results indicated that the model achieved a high human identification accuracy with minimal computational effort and by thus making it a great option for systems that analyze human behavior.