Wi-Fi CSI based Human Activity Monitoring using Machine Learning and Deep Learning algorithms

Monitoring of human activities is a broad field of research. Although there is sensor and vision-based solution is available, however this solution suffers from major limitations. like security and privacy concern as well as relatively expensive components. As a result, in order to alleviate or eliminate these limitations, Radio signal like Wi-Fi is best suitable solution. In order to mitigate above concern, most of the research work are done with the received signal strength indicator (RSSI) methodology. However, accuracy is the one of the major concerns in RSSI. So now a day channel state information (CSI) methodology is one of the popular research areas. However, extracting the Wi-Fi channel state information (CSI) need expensive hardware components like USRP SDR or Network Interface Card (NIC) with PC that make this CSI extraction method too expensive. In our research work we reduce the cost factor by using the low cost hardware called ESP32, through which extracted the channel state information of received signal. Based on the channel state information, detected the human activity like walk, sitting and falling inside a room as well as behind the wall. Since 2.4 GHz WiFi signal can easily penetrate through wall. After extracting the channel state information (CSI), apply the signal processing technique like Wavelet Transformation, Principal component analysis (PCA) in order of reduction of noise, signal transformation and features extraction. Then proposed machine learning algorithm like K-Nearest Neighbor(KNN), Support vector machine (SVM), Decision Tree, Random Forest, and Recurrent Neural Network(RNN) for classification of different activity.