New technology ``WiPose'' that 3D model the posture and movement of a person over the wall with Wi-Fi radio waves appeared



A new technology ' WiPose ' was announced that can detect the posture and movement of a person over the wall with high accuracy using Wi-Fi radio waves. This is expected to overcome the shortcomings of conventional sensing technology using cameras, such as in dark environments or places with obstacles.

Towards 3D Human Pose Construction Using WiFi
(PDF file)

https://cse.buffalo.edu/~lusu/papers/MobiCom2020.pdf

You can see in a single shot what kind of technology 'WiPose' developed by Wenjun Jiang, a researcher at the State University of New York at Buffalo, etc., is seen.

[MobiCom 2020] Towards 3D Human Pose Construction Using WiFi-YouTube


Around the man standing in the center, WiPose's Wi-Fi receiving antennas are separated by a wall.



When a man ran for a running leg in this state, the right 3D model built with WiPose also took the same pose.



Even if you can't see the man behind the camera, you can still see what the man is like in the 3D model.



I also accurately traced the movements that drastically twist my body.



WiPose is a three-step process of 'data collection,' 'data preprocessing,' and 'skeletal model construction.'



First, 'Data collection' uses one Wi-Fi transmitter (transmitter), multiple receiving antennas, and a motion capture system to measure the subject's pose from the reflection of radio waves. Next, we extract

channel state information (CSI) regarding attenuation and diffraction loss of radio signals from the collected Wi-Fi signal data, and perform 'data preprocessing' so that it can be input to deep learning.

Finally, using deep learning, a realistic and natural 'skeleton model construction' is performed. Below is a photo of the subject posing from above, a 3D model created as a model, a 3D model of ' RF-Pose3D ' used for comparison, and a 3D model by WiPose. In the conventional RF-Pose 3D announced in 2018, the elbows of the 3D model are bent unnaturally, the arms that should be raising hands are lowered, the arms are strangely long, The WiPose 3D model is as accurate as a model.



Also, below is a photo of the subject taken in the dark from the left, a 3D model of the model, and a 3D model of WiPose. A normal camera won't help in the dark and you can barely see anything, but WiPose, which uses Wi-Fi radio waves, can build a 3D model without problems.



In a paper that published WiPose, researchers said, ``WiPose enabled us to localize each joint of the human skeleton with an average error of 2.83 cm. It shows a 35% improvement in accuracy compared to the technology.To the best of our knowledge, the study showing that poses can be constructed in 3D using commercially available Wi-Fi devices is This is my first time.'

Regarding the future deployment of WiPose, “Because Wi-Fi devices are familiar, we can easily utilize a wide range of functions with a system like WiPose. For example, privacy of elderly people and patients. Health care technology that does not infringe on humans, games that transfer and pose human poses from a closed environment to virtual space, virtual reality (VR) technology, hide hand movements with clothes and bags etc. Examples include technology that detects shoplifting at end-of-life shopping malls.'

in Software,   Science,   Video, Posted by log1l_ks