Researchers at MIT’s Computer Science and Artificial Intelligence Lab have developed software that uses variations in Wi-Fi signals to recognize human silhouettes through walls. The researchers presented RF-Capture, which tracks the 3D positions of a person’s limbs and body parts even when the person is fully occluded from its sensor, and does so without placing any markers on the subject’s body.
The prototype is sophisticated enough to determine subtle differences in body shapes, and, with 90 percent accuracy, distinguish between 15 different people through a wall. It can even determine a person’s breathing patterns and heart rate.
According to Gizmodo, the RF-Capture could be used to track the movements of an elderly person living alone, and be able to determine if they had fallen down. The technology could also potentially be used in smart homes, if certain gestures detected by the device were used to control appliances. The researchers expect the technology to get more accurate over time.
RF-Capture’s algorithms are implemented in software on an Ubuntu 14.04 computer with an NVIDIA GPU. They implement the hardware control and the initial I/O processing in the driver code of the USRP. The coarse-to-fine algorithm is implemented using CUDA to generate reflection snapshots in real-time. In comparison to C processing, the GPU implementation provides a speedup of 36x.