Fraunhofer Institute for Computer Graphics Research IGDThe growing need for reliable and accurate recognition methods together with recent advances in Deep Learning have fundamentally changed the research field of biometric recognition. It is therefore important to develop efficient biometric solutions that minimize the computational resources required, especially when the biometric methods are to be used on embedded systems or low-end devices.
One possible use case here is head mounted displays – visual output devices such as those employed for VR/AR applications and the Metaverse. Here, for example, the wearer can be uniquely identified from characteristics of the periocular region of the eye, despite high variation with respect to eye position and eye movement. We show a proof of concept in our study.
The demand for ultra-efficient biometric solutions also extends beyond head-mounted displays. Frequently, it is not only the available computing power and memory size that are limiting factors, but also duration, i.e. the execution time of a biometric comparison.
In MixFaceNets, we have succeeded in developing a facial recognition system that can make rapid, highly accurate biometric decisions, despite greatly reduced complexity.