Device-free human identification and device-free gesture recognition
A system can include multiple WiFi-enabled commercial off the shelf (COTS) Internet of Things (IoT) devices disposed within an environment and configured to be a transmitter (TX) or a received (RX) to send or receive data over a WiFi radio frequency communication link. A server can be configured to receive and parse the CSI data transmitted from the RX, store the CSI data with a corresponding human identity label collected for training, train a human identification classifier using a Convex Clustered Concurrent Shapelet Learning (C 3 SL) method, and estimate an identification of a user based on the CSI data and the C 3 SL method. The server can be configured to receive and parse the CSI data transmitted from the RX, transfer the CSI data into real-time CSI frames, store the real-time CSI frames in a database, store the real-time CSI frames with a corresponding gesture label collected in an original environment, and estimate and identify the gesture performed by user using a trained target encoder and source classifier.
1. A system for identifying human gestures in a device-free and privacy-preserving manner in an environment, comprising:
a first WiFi-enabled commercial off the shelf (COTS) Internet of Things (IOT) device disposed within the environment, the first WiFi-enabled COTS IOT device configured to be a transmitter (TX) to send data frames over a WiFi radio frequency communication link;
a second WiFi enabled COTS IOT device disposed within the environment, the second WiFi-enabled COTS IOT device configured to be a receiver (RX) to obtain data frames sent by the TX through the WiFi radio frequency communication link, extract channel state information (CSI) data from the data frames, arrange the CSI data in a predefined format, and transmit the CSI data; and
a server configured to receive and parse the CSI data transmitted from the RX, transfer the CSI data into real-time CSI frames, store the real-time CSI frames in a database, store the real-time CSI frames with a corresponding gesture label collected in an original environment, and use a trained target encoder and a source classifier to estimate and identify a gesture performed by a user, wherein the server is either disposed within the environment or physically remote from the environment, and wherein the server is configured to perform at least one of the following:
train and construct a source encoder and the source classifier; and
train and generate the target encoder by way of unsupervised adversarial domain adaptation, wherein the source encoder is designed as a deep neural network and wherein weights and biases parameters in the network are optimized and updated through backpropagation.
2. The system of claim 1 , wherein the TX is selected from a group consisting of the following: a commodity WiFi router, a smart thermostat, a light switch, a television, and a soundbar.
3. The system of claim 1 , wherein the RX is selected from a group consisting of the following: a commodity WiFi router, a smart thermostat, a light switch, a television, and a soundbar.
4. The system of claim 1 , wherein the RX is configured to perform at least one of the following:
receive the data frames sent by the TX through the WiFi radio frequency communication link;
extract the CSI data from the data frames;
arrange the CSI data in a predefined format; and
send the CSI data to the server.
5. The system of claim 1 , wherein the deep neural network is either or both a convolutional neural network (CNN) and a recurrent neural network (RNN).
6. The system of claim 1 , wherein a gesture recognition is achieved by mapping the real-time CSI frames to a latent feature space and using the source classifier to identify the gesture performed by the user.
7. A system for identifying human gestures in a device-free and privacy-preserving manner in an environment, comprising:
a first WiFi-enabled commercial off the shelf (COTS) Internet of Things (IOT) device disposed within the environment, the first WiFi-enabled COTS IOT device configured to be a transmitter (TX) to send data frames over a WiFi radio frequency communication link;
a second WiFi enabled COTS IOT device disposed within the environment, the second WiFi-enabled COTS IOT device configured to be a receiver (RX) to obtain data frames sent by the TX through the WiFi radio frequency communication link, extract channel state information (CSI) data from the data frames, arrange the CSI data in a predefined format, and transmit the CSI data; and
a server configured to receive and parse the CSI data transmitted from the RX, transfer the CSI data into real-time CSI frames, store the real-time CSI frames in a database, store the real-time CSI frames with a corresponding gesture label collected in an original environment, and use a trained target encoder and a source classifier to estimate and identify a gesture performed by a user, wherein the server is either disposed within the environment or physically remote from the environment, and wherein the server is configured to perform at least one of the following:
train and construct a source encoder and the source classifier; and
train and generate the target encoder by way of unsupervised adversarial domain adaptation, wherein the source classifier is designed as a deep neural network and wherein weights and biases parameters in the network are optimized and updated through backpropagation.
8. The system of claim 7 , wherein the deep neural network is either or both a convolutional neural network (CNN) and a recurrent neural network (RNN).
9. The system of claim 7 , wherein the TX is selected from a group consisting of the following: a commodity WiFi router, a smart thermostat, a light switch, a television, and a soundbar.
10. The system of claim 7 , wherein the RX is selected from a group consisting of the following: a commodity WiFi router, a smart thermostat, a light switch, a television, and a soundbar.
11. The system of claim 7 , wherein the RX is configured to perform at least one of the following:
receive the data frames sent by the TX through the WiFi radio frequency communication link;
extract the CSI data from the data frames;
arrange the CSI data in a predefined format; and
send the CSI data to the server.
12. The system of claim 7 , wherein a gesture recognition is achieved by mapping real-time CSI frames to a latent feature space and using the source classifier to identify the gesture performed by the user.
13. A system for identifying human gestures in a device-free and privacy-preserving manner in an environment, comprising:
a first WiFi-enabled commercial off the shelf (COTS) Internet of Things (IOT) device disposed within the environment, the first WiFi-enabled COTS IOT device configured to be a transmitter (TX) to send data frames over a WiFi radio frequency communication link;
a second WiFi enabled COTS IOT device disposed within the environment, the second WiFi-enabled COTS IOT device configured to be a receiver (RX) to obtain data frames sent by the TX through the WiFi radio frequency communication link, extract channel state information (CSI) data from the data frames, arrange the CSI data in a predefined format, and transmit the CSI data; and
a server configured to receive and parse the CSI data transmitted from the RX, transfer the CSI data into real-time CSI frames, store the real-time CSI frames in a database, store the real-time CSI frames with a corresponding gesture label collected in an original environment, and use a trained target encoder and a source classifier to estimate and identify a gesture performed by a user, wherein the server is either disposed within the environment or physically remote from the environment, and wherein the server is configured to perform at least one of the following:
train and construct a source encoder and the source classifier; and
train and generate the target encoder by way of unsupervised adversarial domain adaptation, wherein the target encoder is designed as a deep neural network and trained by way of unsupervised adversarial domain adaptation to map unlabeled target CSI frames to a shared latent feature space such that a domain discriminator cannot distinguish the domain labels of the data.
14. The system of claim 13 , wherein weights and biases parameters in the target encoder are optimized and updated through backpropagation.
15. The system of claim 13 , wherein the deep neural network is either or both a convolutional neural network (CNN) and a recurrent neural network (RNN).
16. The system of claim 13 , wherein the TX is selected from a group consisting of the following: a commodity WiFi router, a smart thermostat, a light switch, a television, and a soundbar.
17. The system of claim 13 , wherein the RX is selected from a group consisting of the following: a commodity WiFi router, a smart thermostat, a light switch, a television, and a soundbar.
18. The system of claim 13 , wherein the RX is configured to perform at least one of the following:
receive the data frames sent by the TX through the WiFi radio frequency communication link;
extract the CSI data from the data frames;
arrange the CSI data in a predefined format; and
send the CSI data to the server.
19. The system of claim 13 , wherein a gesture recognition is achieved by mapping real-time CSI frames to the shared latent feature space and using the source classifier to identify the gesture performed by the user.