3D action recognition and long-term prediction of human motion
ICVS'08 Proceedings of the 6th international conference on Computer vision systems
Spatio-temporal 3D pose estimation of objects in stereo images
ICVS'08 Proceedings of the 6th international conference on Computer vision systems
Resolving stereo matching errors due to repetitive structures using model information
Pattern Recognition Letters
Hi-index | 0.00 |
In this contribution we introduce the Multiocular Con- tracting Curve Density algorithm (MOCCD), a novel method for fitting a 3D parametric curve. The MOCCD is integrated into a tracking system and its suitability for tracking human body parts in 3D in front of cluttered back- ground is examined. The developed system can be applied to a variety of body parts, as the object model is replaceable in a simple manner. Based on the example of tracking the human hand-forearm limb it is shown that the use of three MOCCD algorithms with three different kinematic models within the system leads to an accurate and temporally sta- ble tracking. All necessary information is obtained from the images, only a coarse initialisation of the model pa- rameters is required. The investigations are performed on 14 real-world test sequences. These contain movements of different hand-forearm configurations in front of a complex cluttered background. We find that the use of three cameras is essential for an accurate and temporally stable system performance since otherwise the pose estimation and track- ing results are strongly affected by the aperture problem. Our best method achieves 95% recognition rate, compared to about 30% for the reference methods of 3D Active Con- tours and a curve model tracked by a Particle Filter. Hence only 5% of the estimated model points exceed a distance of 12 cm with respect to the ground truth, using the proposed method.