18
Jan

Human detection, classification and visualization using Wi-Fi

Remote detection of human activity using non-intrusive observation methods is an important technological need, particularly for the security industry.  Modern Smart Cities ideas also require new and novel methods of detecting and classifying human movement and general activity. Typically observation/detection is accomplished through digital or analog video cameras. But cameras have a limited field of view which allows avoidance. Deploying many cameras with overlapping coverage may be prohibitively expensive to solve this problem, particularly covering large areas. Special night-vision cameras are also needed for night time detection.  Wi-Fi sensing is an alternative solution to this problem since Wi-Fi signal already exist in an environment to be monitored or may be easily added with a benefit of providing device and human network connectivity.

The following video, provided by our sponsored researches Vahid and Hadi, demonstrates the results of a Proof of Concepts deployment and test. In this video 3 panes are shown. The left most pane shows how the real-time CSI vector changes as captured from a Wi-Fi access point. The middle pane shows an actual person moving in the Wi-Fi field recorded by a regular digital camera. The right most pane shows a reconstructed image of a person as derived from the changes in the CSI vector and overlaid onto the static environment image, for easy visualization.

[ CSI Vector Captures ]         [ Video Capture ]         [ Wi-Fi Detection ]

Previous implementations of Wireless sensing use Wi-Fi received signal strength indicator (RSSI) signals for probing the environment. Modern Wi-Fi standards with OFDM and MIMO technologies, which are adopted in 802.11 n/ac, provide new environment monitoring capabilities at the Wi-Fi receiver side using Channel State Information (CSI). In an empty environment with just a Wi-Fi transmitter and a receiver, the received signal is identical to the transmitted signal. However,  when there are objects in the environment (moving or stationary), the receiver gets multiple copies of the transmitted signal and therefore the received signal will be a perturbed copy of the transmitted signal. This perturbation is not constant over different frequencies and depends on the locations, build-materials, and speeds of the object traversing the environment. Channel State Information (CSI) is a complex-valued vector that quantifies this effect on each of the Wi-Fi frequency bands.

SignalDiagram

CSI carries new information which allows much more precise detection and classification of objects. CSI is captured by measuring the received signal strength and phase across multiple Wi-Fi subcarriers – up to 114 on a 40 Mhz channel.  CSI vector is then constructed which captures both signal strength and phase information for OFDM sub-carriers and between each pair of transmit-receive antennas receiving multiple signal copies.  As objects move through the monitored Wi-Fi field, CSI vector changes in real-time in a specific way.

Several open source projects allow convenient capture methods for real-time CSI data from Intel Wi-Fi chipsets and from Qualcomm/Atheros chipsets.  Futurebound Corp. provided equipment and logistical support to our affiliated researcher Vahid Pourahmadi and his colleague, Mohammad Hadi Kefayati, to further their efforts in developing a proof of concept system for human detection and visualization of such detection using Wi-Fi and CSI methodology. More details can be found at https://arxiv.org/pdf/2001.05842.pdf