3D facial recognition

Poorly-lit areas have historically created an impediment to surveillance operations using face recognition technologies. But that is changing with the development of new face recognition technology that works just as efficiently in darkness.

Previously, face recognition systems were only useful in conditions where enough light was available or where shadows were minimal. But the new system, created by a team of researchers at the Karlsruhe Institute of Technology in Germany, uses a sophisticated technique wherein multiple infrared images of a person’s face are compared by a computer to images taken in daylight. These advanced comparisons are conducted via a computer program based on the deep neural network – a system that can imitate the function of the human brain.

According to the team of researchers, 4,585 images were analysed by the deep neural network. These images were taken in both visible light and infrared. Based on this information, the software was able to display a match in less than a second – 35 milliseconds to be precise.

Using this advanced technique, the researchers claim that they have been able to advance the state of the art of facial recognition technology by about 10%. Although the accuracy rates of the software are yet to reach optimal levels, they are reasonable for a first attempt. The 4,585 images were of 82 people. The computer that processed the images was relatively fast and the rate of accuracy was about 80% that of images captured in well-lit conditions. However, in conditions where only a single visible light image was present, the accuracy of the system slid to about 55%. The researchers are confident that the accuracy rate can be improved, as eventually larger datasets become available and the interconnectivity of networks improves.