This project was able to identify the 4 individuals and the empty room with 86 percent accuracy using only the channel state information of a Nexus 5 phone that had the Nexmon Firmware patch. A router was set in a specific configuration as to not hop channels (for ease of use) and then it sent out beaconing packets. Individuals walked one at a time past the cell phone (Nexus 5).
The data from the Nexus 5 was a .pcap file (see the pcap-data folder). This file was put into the csi_parser.py script to convert it to a .csv. In the original experiment the csv underwent normalization and was changed into an image in Matlab. However, Tensorflow/Keras can handle .csv files directly and can do normalization itself.
Our findings seem to indicate that the Neural Network identifies humans based on Gait.
Here is the Neural Network that was most effective.
More information in the .pdf inside the paper folder
Each person walked for 30-40 minutes over all (After the data was gathered and fed into the NN this is clearly more then nessisary for 1 room of data, instead of sending in 30 seconds per frame it is likely the same results can be done with 5-8 seconds) with 5-7 minutes per method (multiple walking paths past the router cell phone for the fast/slow walks).