Radar Love

Deep learning and high frequency radio signals work together to determine if someone is feeling happy, or if they giga-hurts.

Nick Bild
a month agoMachine Learning & AI
Experimental setup (📷: A. Khan et al.)

Detection of emotional states has numerous practical applications in areas as diverse as medical diagnosis, education, fraud detection, and video gaming. The core sensing technologies typically employed include cameras or sensors attached to the body that collect physiological parameters. These types of solutions limit the applications in which the technologies can be used — cameras present privacy concerns to users, while on-body sensors can be cumbersome and uncomfortable for long-term use.

The problem of emotion detection was approached from a new angle by researchers at Queen Mary University of London. They have recently reported their method that makes use of non-intrusive wireless radio frequency (RF) signals and a deep learning framework for emotion detection.

The experimental setup was assembled in an anechoic chamber to eliminate interfering electromagnetic waves. Vivaldi type antennas operating at 5.8 GHz formed the radar, with one antenna transmitting signals towards the body, while another antenna received the reflected signals.

The team recruited a cohort of fifteen participants that were shown four different video clips intended to elicit particular emotional responses. RF signals were reflected off of the participants as they watched the videos and, after filtering, these signals were used to infer heartbeat and breathing metrics. This data was fed into a custom convolutional neural network with long short-term memory sequence learning cells that correlated these metrics with the emotional state of the subject.

The model is currently able to decipher the emotions of anger, sadness, joy, and pleasure. The network achieved state-of-the-art classification accuracy when compared with several other traditional machine learning algorithms, such as Random Forests and Support Vector Machines. The team’s new method achieved a classification accuracy of 71.67%.

The researchers believe that their new technique will be widely adopted in research applications in the future due to the low-cost, high accuracy, and the non-invasive nature of the approach.

Nick Bild
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.
Related articles
Sponsored articles
Related articles