Ground Truth Extraction


In the current section, we describe the camera setup and the marker-based procedure for ground truth extraction.

In order to obtain the ground truth, we use a visual marker-based acquisition technique within a two-camera setup. Prominent color markers are attached on both sides of the user's body, corresponding to the underlying joints, namely the shoulders, the elbows, the wrists, the hips, the knees and the ankles. Semi-automatic annotation on the RGB-D sequences leads to the computation of the 3D position of each joint.

The second (rear) camera is placed behind of the user and slightly elevated in order to avoid interference between the two cameras. The two-camera setup guarantees that all markers are visible by at least one of the cameras at all times and, therefore, facilitate ground truth information provision even for occluded joints. It is worthy mentioning that the rear camera is only used for this purpose -i.e. ground truth computation of occluded joints for certain frames- and, threfore, the corresponding image sequences are not provided.