Summary of 0.
Face pose estimation algorithm is mainly used to estimate 3d Euler Angle of input face block. The general reference system is the camera coordinate system, that is, the camera is selected as the origin of coordinates. Pose estimation can be used in many business scenarios. For example, in face recognition system, pose estimation can assist in the screening of input samples (generally, faces must be recalled and recognized relative to positive faces). Face pose algorithm is also indispensable in some scenarios that require face orientation as an important basis for business, such as driver’s left and right look detection in fatigued driving products.
Face pose estimation algorithm, which can be roughly divided into two categories: One is through 2D calibration information to estimate 3D attitude information algorithm, such as the first calculation of the key points of the face, and then select a reference frame (the key points of the average positive face), calculate the transformation matrix of the key points and the reference frame, and then through the iterative optimization algorithm to estimate the face attitude (can refer to the SolvePnP algorithm in Opencv, Or 1 and 2 in reference); The other is to train a regressor in a data-driven way to make a direct prediction of the blocks of input faces.
Both methods have advantages and disadvantages and can be applied to different degrees in projects. Let’s take a look at the comparison of advantages and disadvantages of the above two methods. For the convenience of distinguishing, we call the algorithm of the first type landmark-based and the algorithm of the second type landmark-free.
Compare the two kinds of methods
1. Introduction to Method 1