Computer scientists in Germany have developed a new infrared facial recognition system that uses thermal images to identify people in the dark.
The system uses multiple infrared images to create a unique “thermal signature” for a person. This thermal signature is based on heat maps of their face and it can be used to identify people in poor lighting or even total darkness by comparing infrared photos to visible light photos of the same person.
The system uses an advanced “deep neural network” program that can take infrared images of people and match them with visible light photographs and is reportedly 10% more reliable than any identification system currently in use by police.
Deep neural networks are complex software programs that can be taught to recognize people and objects and make other connections in ways that are similar to how the human mind works.
By collecting multiple regular photos of people with different facial expressions taken in different lighting levels, the researchers were able to “train” the deep neural network to recognize people in infrared images.
In addition to being more reliable, this infrared facial recognition system is fast, with the ability to identify a person in just 35 milliseconds.
Current facial recognition technology relies heavily on comparing well-defined photos of people in good lighting. This type of facial recognition performs poorly when trying to identify people in pictures with poor lighting and it doesn’t work at all in total darkness.
This new identification system is still not ready for use by police. The system has an 80% success rate in identifying people when it has multiple photos of the person to compare the infrared image to. However, the success rate drops to 55% when the system only has one photo of the person to compare the infrared image to.
The researchers believe that this new identification technology shows real promise for use in night-time or poorly-lit surveillance operations.