SASE: RGB-Depth Database for Human Head Pose Estimation

摘要

Head pose estimation has become very important in relation to facial and emotional recognition, as well as in human-computer interaction. There is an ultimate need for a 3D head pose database in order to develop head pose estimation methods using RGB and depth information. There are a few available datasets, such as Biwi Kinect head pose datar base, which is composed using Kinect 1, but it offers low-quality depth information. In this paper, a new 3D head database, SASE, is introduced. The data in SASE is acquired with Microsoft Kinect 2 camera, including RGB and depth information. The SASE database is composed by a total of 30000 frames with annotated markers. The samples include 32 male and 18 female subjects. For each person a large sample of head poses are included, within the bounds of yaw from -45 to 45, pitch -75 to 75 and roll -45 to 45° of rotation around each axis. The details of acquiring the database and its characteristics are explained in detail.