Design Of Fast Three-dimensional Fingerprint Acquisition System Based On Structured Light

With the development of economy and the advancement of science and technology,information security has become more and more concerned by people.Traditional passwords and keys are no longer suitable for all occasions.Driven by the increasing requirements of information security,biometrics technology has become a breakthrough point in the field of mobile payment and public safety.Among all the technologies,the fingerprint recognition technology is the most widely used and accepted by people.However,the traditional fingerprint collection technology still has some defects.The three-dimensional fingerprints have no deformation of the fingerprint,contain more feature information,and have no public health risks.The research of 3D fingerprint collection has become a new development direction and breakthrough for fingerprint collection technology.Structured light 3D measurement technology has been introduced into the field of fingerprint recognition due to its advantages such as high accuracy and wide adaptability.In this paper,the hardware is designed for 3D fingerprint measurement with small area and high depth accuracy.The projector with a small volume and a small projection ratio and a high frame rate camera are selected to build a structured light 3D fingerprint acquisition system.Based on the hardware design,the software design from system calibration to 3D reconstruction is implemented,which mainly includes background segmentation,region of interest extraction,phase extraction,and multifrequency unwrapping in the time domain.Finally,the 3D fingerprint is reconstructed.The phase shift method needs to project multiple fringe patterns during measurement and the time-domain unwrapping method needs to project fringe patterns with multiple frequencies,resulting in a large number of fringes and a slow measurement speed.A phase estimation method from single fringe pattern based on generative adversarial networks(GAN)is proposed.The feature of single fringe pattern is extracted by convolution neural network.The phase value is then extracted by fitting the input feature into the GAN network.Consequently,the number of fringe patterns is reduced from 3-8 in original system to 1 for each frequency.the total number of fringe patterns in the system is greatly reduced,thereby the measurement is speedup and the risk of finger vibrations during measurement is reduced.Thus,fast collection of 3D fingerprints is realized.