On basis of binocular stereo vision, the paper puts forward a method to acquire the 3D data information of human face feature points, that is, by using a nominal camera to shoot 3D image pair and the method with the combination of prior experience in local face feature and Canny edge detection, 15 feature points are respectively extracted from the planar image. Under the conditions of epipolar constraint, uniqueness constraint and order consistency, the left and right image pair is 3D matched, the parallax is calculated, and the 3D spatial data of the feature points is synthesized. The experimental results show that: the synthesized 3D data information of the feature points is in line with the actual situation; the relative error is less than 4%.