Anatomical landmark detection on 3D human shapes by hierarchically utilizing multiple shape features

The availability of 3D human body shapes enables applications such as digital anthropometry by exploiting the geometric information of 3D shapes. In this paper, we propose a method for detecting anatomical landmarks on 3D human shapes by hierarchically utilizing multiple shape features. The backbone of our method is to compute dense correspondences between a pair of template and target shape, and the detection is achieved by transferring the annotated landmarks of the template shape to the target shape. We also investigate several techniques to further enhance the detecting accuracy, such as template selection, fine search and late fusion. Multiple kinds of shape features are used in different parts of our method, and each of them contributes to the improvement of detection accuracy. In experiments, we validate the effectiveness of each part in the proposed method. And our method also demonstrates the state-of-the-art performance in terms of the average detection accuracy.