Multiview Video Datasets for Evaluation and Testing of 3d Tracking and 3d Reconstruction Algorithms

This paper presents a dataset that can be used for testing and evaluation of computer vision-based algorithms for 3D reconstruction and 3D person tracking. Additional possible uses include the building of 3D human head models and the production of stereoscopic sequences. A number of different scenes are included in the dataset. They are mostly single-subject scenes captured with two different lighting conditions (optimal and sub-optimal), subject motion based on simple as well as random motion trajectories, different distances from the camera and occlusion. The dataset incorporates video data captured simultaneously from a system of 2 cameras, each employing three sensors. Additionally, depth data originating from the system itself is available.