3D emotion recognition hardware gets better, but the software might still be iffy
A South Korean research team is boasting a 3D emotion-recognition innovation. It claims to have married a tiny, specialized camera with sophisticated facial recognition algorithms to detect faces and the emotions flitting across them.
The scientists, from the Korea Advanced Institute of Science and Technology, or Kaist, say the camera is a light-field device small enough to be installed in some phones. With its micro-lens array, the camera can detect special and directional information for each image.
Depth is plumbed using a vertical cavity, surface-emitting laser in the camera working in near-infrared light, a task conventionally handled by reading environmental light on a subject alone, according to Kaist’s Department of Bio and Brain Engineering researchers.
Three-dimensional IR systems are better than two-dimensional visible light in discerning facial changes related to each emotion, according to the researchers.
The depth map built by the laser and a filter reportedly cut image-reconstruction errors by up to 54 percent regardless of how visible light was used to illuminate a subject, they said.
That is the technology that gets images to device sensors where facial recognition software analyzes the data.
The other half of the research’s outcome is more controversial.
AI software used in the team’s experiments recognized and categorized four emotions: disgust, sadness, happiness and anger with what was judged in the team’s paper as an “exceptional” average success rate of 0.85.
Thirty-two adults were instructed to express a facial expression in an effort to squelch gender, age and ethnicity variables. The result was 281 raw images to which were added images marred by noise.
Similar systems could be used in health care to help diagnose problems, improve social perception and make human-machine interaction safer and more efficient.
Some in the facial recognition and other industries feel expectations for emotion recognition and the actual capabilities are at odds.
That is not unusual with new technology, but in this case, governments and businesses are deploying systems that could have dramatic, negative effects on individuals if emotion analysis is overpromised.
Yet industry and human rights criticism has not turned off entrepreneurs and funders.