FEATURE-POINT TRACKING METHOD AND FEATURE-POINT TRACKING DEVICE

摘要

To improve the robustness of feature-point tracking by predicting the movement of feature points in time-series image frames with a high accuracy. A feature-point tracking device (10) detects the three-dimensional positions (P0, P1, P2) of the three feature points of an object person (A) on a time-series basis, calculates the rotational angles (α, β, γ) and displacement of a face coordinate system from a camera coordinate system on the basis of the three-dimensional positions (P0, P1, P2) at the past imaging timing, predicts the rotational angles (αp, βp, γp) and displacement of a face coordinate system at the current imaging timing on the basis of the rotational angles (α, β, γ) and displacement of the face coordinate system from the camera coordinate system, and calculates the three-dimensional prediction position (Pn1) of a feature point at the current imaging timing on the basis of the rotational angles (αp, βp, γp) and the displacement. When detecting the positions of feature points, the feature-point tracking device (10) sets windows on the object image frames on the basis of the three-dimensional prediction position (Pn1) and detects the feature points with respective windows as objects.