In medicine, tracking an object in 3D-space is important for disease diagnosis. This paper presents a method for
a 3D structure reconstruction of dynamic facial key points for diagnosis of facial paralysis. In our method, a synchronous multi-angle high-speed capture system (SMAHCS) was first set up to capture patient’s multi-view images. Then, a camera self-calibration technique is used to estimate cameras’ parameters. After that, the key
points of frontal image are initialized. Next, the corresponding points of these key points in other views are
searched by using an epipolar line constraint and a coarse-to-fine matching process. Finally, these dynamic key
points are reconstructed in 3D-space. Using epipolar line constraint, the time-consuming process of matching is
reduced. Hence, the movement of dynamic 3D key points can be tracked and then the dynamic 3D facial structure
is reconstructed and used for dynamic 3D-feature extraction. Besides the diagnosis of facial paralysis, our
method can be also used for applications in which the dynamic structures of objects are required.