Abstract:Human activity recognition and person identification using Channel State Information (CSI), according to IEEE 802.11n standard WiFi signals, has recently gained much attention. The WiFi signals captured at the receiver contain fine-grained information about the sub-carriers, which embeds information about the person and their activity. Identifying the person in a smart home environment is as important as identifying the activity. To fulfill this requirement, in this work, we propose three machine-learning based models; (1) a single convolutional neural network-based model to classify the person and activity together accurately, (2) a two-phase deep learning model that uses a convolutional neural network where the person is classified first and based on the identified person, the activity is classified using another trained model, and (3) a two-phase multi-stage deep neural network model to minimize false negatives. This multi-stage model uses transfer learning-based popular networks like ALexnet, Googlenet, and Squeezenet to minimize false negatives. The multi-stage phase is stacked with the pre-trained models based on their accuracies. The first and second models are proposed to increase the system's accuracy in identifying the person and the activity. The third model is proposed to reduce false negatives to avoid missing critical activities corresponding to key persons. Our proposed models can achieve an accuracy of 81.4% to 83.9%. Our proposed model is also able to address the modified objective of minimizing the false negatives with an acceptable increase in false positives. In addition to the existing benchmark dataset, experiments are conducted with three more participants in the laboratory environment, and CSI data are collected. We used the existing dataset to train and validate the system and tested the system with our experiment dataset. The test results are found to be approximately 90% accurate in detecting the intruder.

WiPg: Contactless Action Recognition Using Ambient Wi-Fi Signals

WiDIGR: Direction-Independent Gait Recognition System Using Commercial Wi-Fi Devices

WiFine: Real-Time Gesture Recognition Using Wi-Fi with Edge Intelligence

WiWalk: Gait-Based Dual-User Identification Using WiFi Device.

Wi-Motion: A Robust Human Activity Recognition Using WiFi Signals

Wi-PIGR: Path Independent Gait Recognition with Commodity Wi-Fi

An improved Wi-Fi sensing-based human activity recognition using multi-stage deep learning model

Wi-Exercise: An Indoor Human Movement Detection Method Based on Bidirectional LSTM Attention

Towards Position-Independent Sensing for Gesture Recognition with Wi-Fi.

Human Activity and Gesture Recognition Based on WiFi Using Deep Convolutional Neural Networks

CSI-F: A Human Motion Recognition Method Based on Channel-State-Information Signal Feature Fusion

A Survey of Human Activity Recognition Using WiFi CSI

Location- and Person-Independent Activity Recognition with WiFi, Deep Neural Networks, and Reinforcement Learning

WiGNN: WiFi-Based Cross-Domain Gesture Recognition Inspired by Dynamic Topology Structure

Wi-AM: Enabling Cross-Domain Gesture Recognition with Commodity Wi-Fi

WiHF: Enable User Identified Gesture Recognition with WiFi

Understanding And Modeling Of Wifi Signal Based Human Activity Recognition

Vi-WiFi-Gate: a WiFi sensing gait recognition method inspired by the image vision field

We Hear Your Activities through Wi-Fi Signals

WiFi-based Cross-Domain Gesture Recognition via Modified Prototypical Networks

Wi-Fi sensing based person identification and activity recognition using two-phase deep learning model