Three-dimensional face acquisition method and device based on trinocular camera

Abstract

The invention discloses a three-dimensional face acquisition method and device based on a trinocular camera. The method includes: projecting the face key points into the left IR speckle texture face image and the right IR speckle texture face image; calculating coordinates of corresponding face key points in the left IR speckle texture face image and the right IR speckle texture face image, iterating projection positions of the face key points on the left IR speckle texture face image and the right IR speckle texture face image, correcting errors, finally traversing the whole RGB face image toobtain an accurate face depth map, and obtaining a final face point cloud by utilizing internal reference matrix back projection. The binocular stereo matching algorithm is enabled to perform high-precision matching by utilizing normal information so that the influence of human face curved surface affine transformation can be eliminated and the depth precision can be enhanced; through the face depth map with reduced resolution, the search range of the stereo matching parallax of the original resolution in the epipolar direction is constrained, the calculation complexity is greatly reduced, and the algorithm calculation efficiency is improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, in particular to a three-dimensional face acquisition method and device based on a trinocular camera. Background technique [0002] In recent years, with the rapid development of artificial intelligence technology and the continuous improvement of computer computing power, face recognition technology, as a typical artificial intelligence technology, is being increasingly used in electronic payment, identification, criminal investigation, Security control and other fields, and show great advantages in these fields. However, it is undeniable that most of the current face recognition algorithms are 2D face recognition based on plane images. And texture information features will inevitably be lost, which also leads to the limitation of face recognition accuracy. [0003] With the rapid development of 3D reconstruction technology, the face recognition algorithm is transitioning from 2D face ...

AI Technical Summary

Problems solved by technology

Especially when the speckle particles projected by the speckle device are large and the density is low, and the angle between the face and the optical axis of the camera is large, the same speckle on the face surface will appear in the face images captured by the left and right cameras. , there are obvious affine transformation differences in the shape of speckle particles, these differences will lead to a large accuracy error in the depth map obtained by the binocular stereo matching algorithm, and then affect the accuracy of face point cloud and face 3D reconstruction

At the same time, since the stereo matching algorithm needs to do a pixel-by-pixel matching search in the epipolar direction, when the face is close, the parallax search range is large, and the binocular stereo matching algorithm has high computational complexity, which leads to time-consuming calculations. longer

Images