An Artificial Intelligence-Driven Framework for Real-Time CHPD Using Wi-Fi Channel State Information

In the quiet corners of our homes and offices, invisible waves of Wi-Fi are constantly bouncing off walls, furniture, and the air itself. These waves carry our emails and videos, but they also carry a hidden story about the space they travel through. When a person moves through a room, their body subtly disturbs these waves, changing how they arrive at a receiver. This disturbance is known as Channel State Information, a detailed map of the signal's journey that reveals not just the strength of the connection, but the specific shape and timing of every ripple. While cameras have long been the standard for watching who is in a room, they raise concerns about privacy and require expensive hardware. Researchers have begun to ask if the Wi-Fi signals already present in our lives could serve as a silent, invisible observer, capable of counting people and detecting movement without ever taking a picture.

A researcher has now built a system that turns this idea into a working reality. They created a framework that uses two small, inexpensive computer chips to listen to the Wi-Fi signals in a room and determine exactly how many people are present, what they are doing, and whether they are sitting, standing, or walking. The setup is simple: one chip acts as a transmitter, sending out a steady stream of Wi-Fi packets, while a second chip acts as a receiver, catching those packets as they arrive. When the room is empty, the packets arrive in a predictable pattern. But when a person enters the space, their body reflects and scatters the signals, creating a unique fingerprint of change in the data. The researcher captured these changes, converting the raw signal variations into a stream of numbers that a computer could understand.

To make sense of this complex data, the researcher developed a specialized computer brain, a type of artificial intelligence designed to learn from patterns. They did not rely on a single method but combined three different techniques to mimic how a human might analyze a scene. First, the system looked at the spatial layout of the signal, identifying which parts of the Wi-Fi spectrum were most affected by a person's presence. Next, it examined the sequence of events over time, tracking how the signal changed from one moment to the next to understand movement. Finally, the system learned to focus its attention on the most important parts of the data, ignoring the background noise that often plagues wireless signals. This hybrid approach allowed the computer to distinguish between a room with no one in it, a room with one person, and a room crowded with up to five people or even a larger group.

The researcher tested their system by recording thousands of signal samples in a controlled environment. They labeled the data according to the ground truth of the room: empty, one person, two people, up to five people, a crowd, and specific activities like sitting, standing, or walking. They fed this data into their artificial intelligence model, which learned to associate specific signal patterns with these specific conditions. The results were promising. The system successfully classified these ten different scenarios with a test accuracy of 78.57 percent. While not perfect, the performance was strong enough to demonstrate that the technology works. The researcher noted that the system sometimes confused the number of people when the difference was small, such as mistaking one person for five, likely because the signal changes in those specific conditions can look quite similar. However, the overall ability to detect presence and activity was robust, even when the people were not directly visible to the sensors, proving that the system works even when the line of sight is blocked.

To bring this technology to life, the researcher built a real-time dashboard that displays what the system is seeing. This interface shows the live signal strength, the number of people detected, the predicted activity, and the confidence the computer has in its answer. It also tracks environmental factors like temperature and humidity, which can influence the signals, ensuring the system remains aware of its surroundings. The entire process happens in real-time, with the system processing twenty consecutive frames of data at once to make a decision. This approach allows the system to react quickly to changes in the room, providing a continuous stream of information about occupancy without ever needing a camera lens.

The significance of this work lies in its potential to offer a privacy-preserving alternative to traditional surveillance. Because the system relies on the existing Wi-Fi infrastructure and does not capture images or video, it can monitor spaces without invading personal privacy. It uses low-cost hardware that is widely available, making it accessible for smart homes, healthcare monitoring for the elderly, and security applications where cameras are not suitable. The researcher acknowledges that this is a step forward, but not the final destination. They plan to improve the accuracy by gathering more diverse data and refining the artificial intelligence model. For now, they have demonstrated that the invisible waves of Wi-Fi can indeed tell us who is in the room, offering a new way to sense our world that is both effective and respectful of our privacy.