The recent availability of compact and inexpensive depth cameras or the combination of high frame-rate cameras aboard the last generation of smartphones and tablets with computer vision techniques is enabling new applicative scenarios for 3D biometrics. In this chapter, new methods for performing real time 3D face and ear acquisition for biometric purposes, by means of non-specialized devices coupled with computer vision algorithms, are presented and discussed. To this regard, algorithms addressing the Synchronous Location and Mapping (SLAM) problem, like Parallel Tracking and Mapping (PTAM) and Dynamic Tracking and Mapping (DTAM) or the more recent MonoFusion and MobileFusion could make 3D biometrics much more exploitable than in the past, allowing even a not trained user to capture 3D facial features on the fly for improved recognition accuracy and/or reliability. This chapter also highlights the relevance of an efficient implementation of these inherently computationally intensive techniques, which can greatly benefit from GPU-based parallel processing, in making these solutions approachable in real-time on mobile hardware architectures. To this aim, the robustness of recognition algorithms to lower resolution, noisy face geometry, would result in a very desirable feature.
Real Time 3D Face-Ear Recognition on Mobile Devices: New Scenarios for 3D Biometrics “in-the-Wild”
Ricciardi S.;
2017-01-01
Abstract
The recent availability of compact and inexpensive depth cameras or the combination of high frame-rate cameras aboard the last generation of smartphones and tablets with computer vision techniques is enabling new applicative scenarios for 3D biometrics. In this chapter, new methods for performing real time 3D face and ear acquisition for biometric purposes, by means of non-specialized devices coupled with computer vision algorithms, are presented and discussed. To this regard, algorithms addressing the Synchronous Location and Mapping (SLAM) problem, like Parallel Tracking and Mapping (PTAM) and Dynamic Tracking and Mapping (DTAM) or the more recent MonoFusion and MobileFusion could make 3D biometrics much more exploitable than in the past, allowing even a not trained user to capture 3D facial features on the fly for improved recognition accuracy and/or reliability. This chapter also highlights the relevance of an efficient implementation of these inherently computationally intensive techniques, which can greatly benefit from GPU-based parallel processing, in making these solutions approachable in real-time on mobile hardware architectures. To this aim, the robustness of recognition algorithms to lower resolution, noisy face geometry, would result in a very desirable feature.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
