Multiple Human Pose Estimation (3D Pictorial Structures)
We introduce two challenging 3D human pose datasets for multiple human 3D pose estimation. This work has been presented in CVPR 2014. The presentation is available here: Download Slides.
The talk from CVPR 2014 is available online : Watch Talk.
Abstract:
We address the problem of 3D pose estimation of multiple humans from multiple views. This is a more challenging problem than single human 3D pose estimation due to the much larger state space, partial occlusions as well as across view ambiguities when not knowing the identity of the humans in advance. To address these problems, we first create a reduced state space by triangulation of
corresponding body joints obtained from part detectors in pairs of camera views. In order to resolve the ambiguities of wrong and mixed body parts of multiple humans after triangulation and also those coming from false positive body part detections, we introduce a novel 3D pictorial structures (3DPS) model. Our model infers 3D human body configurations from our reduced state space. The 3DPS model is generic and applicable to both single and multiple human pose estimation. In order to compare to the state-of-the art, we first evaluate our method on single human 3D pose estimation on Human Eva-I [22] and KTH Multiview Football Dataset II [8] datasets. Then, we introduce and evaluate our method on two datasets for multiple human 3D pose estimation.
This work is the result of our collaboration with the Computer Vision and Multimodal Computing group.
This work was funded in part by the DFG DACH Project “Advanced Learning for Tracking and Detection in Medical Workflow Analysis”.
Contact Person: Vasileios Belagiannis
Shelf Dataset
Download Shelf (~16GB)
We have annotated the body joints of four actors using the cameras 2, 3, and 4. We have performed triangulation using the three camera views for deriving the 3D ground-truth. The actor 4 (Vasilis) is occluded in most of the camera views and thus excluded from the evaluation.
Evaluation frames: [300 - 600].
There is annotation for more frames which can be used for training.
Campus Dataset
Download Campus (~1GB)
We have annotated the body joints of the main three actors for the frames that are observed from the first two cameras. The ground-truth for the third camera view is the result of the triangulation (between camera 1 and 2) and then projected to camera 3. The original dataset (campus, sequence 1) is available at: CVlab page.
Evaluation frames: [350 - 470] and [650 - 750].
There is annotation for more frames which can be used for training.
