No abstract

No abstract

No abstract

No abstract

This article presents a method for creating ground truth data for 3D head pose estimation utilizing a Vicon MX motion capture system. In the context of this work 22 video sequences of six individuals were captured using an off-the-shelf RGB camera with a rate of 30 frames per second. The sequences consist of scenarios that show head rotations with respect to each rotational axis undergoing different lighting conditions. The video data was captured at a resolution of 800 × 600 pixels in the RGB24 format, allowing for further image processing to detect characteristic features. In addition to the provided 3D head pose data, information about the gaze was recorded by requesting the user to follow points displayed on a screen. The article describes the measurement setup as well as the experimental procedure in detail. The goal is to show how the 3D trajectories of the markers of the motion capture system can be transformed into reference data for the state vector of the head. Furthermore, a dataset created with the proposed method is provided that allows a validation of appearance-based head pose estimation algorithms.