researchers in Carnegie Mellon University I’m testing a system that uses Wi-Fi signals to determine a person’s position and pose in a room. In our tests, we placed his regular Wi-Fi routers, specifically the TP-Link Archer A7 AC1750 devices, at each end of the room, with varying numbers of people in the room. AI-powered algorithms analyzed interference in Wi-Fi signals generated by people.
Wireframe images generated from Wi-Fi monitoring looked pretty accurate most of the time. claim That estimate is as good as some “image-based approaches”. Using Wi-Fi through your camera also has some advantages and attractions. First, wireframe estimation of human poses is more respectful of human privacy. Second, Wi-Fi-based perception does not require light and can detect body pose even when there are objects in the way that would obscure a conventional camera’s view. Another big draw to the find is that the Wi-Fi routers used are cheap at just $30 each, making them much more accessible than expensive and power-hungry solutions like Radar and his LiDAR. am.
image 1 of 2
You can see the series of synced images above. Video still on the left, Wi-Fi sensed wireframe generated by AI on the right. The number, location and pose detection of people seems very accurate. A paper published by researchers at Carnegie Mellon University provides detailed background on how this was done. In a nutshell, the Wi-Fi-based recognition technology described here is based on the channel state information (CSI) of the Wi-Fi signal. It represents the ratio of the transmitted signal wave to the received signal wave. This data is processed using a computer vision-savvy neural network architecture capable of performing dense pose estimation. To simplify and speed up the generation of wireframe-style human representations, the researchers divided the human form into 24 segments.
The researchers admit that the methods outlined in detecting humans and their positions/poses are not without problems, and that some obvious errors are still seen in test scenarios. You have kindly provided us with some comparative images showing an example. This can be attributed to issues such as the human being in unusual poses and having too many subjects in the room at once (the engine optimally supports no more than three individuals in her).
There’s still a lot of work to be done, and the researchers suggest that the technology outlined could be improved in many ways, but it’s primarily the Wi-Fi-based derived from better public training data for the perception of , some may worry about the new threat of Wi-Fi routers spying on them.
