
This technology could let you see through walls like Superman.
Researchers at the Computer Science & Artificial Intelligence Laboratory at Massachusetts Institute of Technology have developed RF-Capture, a compact antennae array that uses multiple snapshots captured by wireless signals to create one single image of a person obscured behind a wall.
The wireless signals reflect off of the target person and back to the machine, and are able to capture whether a person is moving, how they are moving, and even trace their handwriting. A reconstruction algorithm is used to put together the many different snapshots taken by the device. Using silhouettes, the device can also distinguish between people, and remember and identify them by using a classifier. In testing, it had 95.7 percent accuracy when distinguishing between five people, or 88.2 percent accuracy when distinguishing between 15 users.
The device is based on existent motion detectors using RF signals, but those abstracted the entire human body into a single point of light instead of being able to visualize specific body parts. Earlier devices ran into trouble because different body parts reflect RF signals differently; a limb might deflect the signal away from the sensor, for example. Previous sensors lacked the semantics to understand what part of a figure they were viewing.
Two algorithms in the RF-Capture get around this problem. One, a coarse-to-fine algorithm, scans the space around a figure and detects the RF reflections of various human limbs. A second algorithm stitches the many snapshots generated by the first step into a complete form.
The RF-Capture was tested on 15 individuals and compared to the readings from a Kinect sensor. Like the Kinect, the RF-Capture can capture data from only a certain field of view in front of it. The researchers believe that a wider field of view and complete skeletal tracking could be added as the understanding of wireless reflections and how they interact with computer graphics increases.
The researchers’ paper about the device will be published in SIGGRAPH Asia 2015.