Citation Context

...n is still too expensive for large data sets and skeletons with many degrees of freedom. Since articulated models are not very realistic models of the human body, implicit surfaces based on metaballs =-=[21]-=-, shape-from-silhouette model acquisition [8], or the learned SCAPE body model [1, 3] have been proposed. While these approaches model the human body without clothing, Balan and Black [2] have used SC...

Citation Context

...lar algorithms can be used to track tissue in medical data [16]. Passive methods can also be employed to jointly capture kinematic motion parameters and surface deformations of tightly dressed humans =-=[5, 22]-=-. None of these algorithms, though, can track arbitrarily dressed people at a level of accuracy comparable to ours. Statistical models have also shown their potential to track confined deformable surf...

Citation Context

... hulls, point clouds and voxel representations (not considered further here); 2) part-based models using generic shape primitives such as cylinders or cones [9], superquadrics [14, 22] or “metaballs” =-=[17]-=-; 3) humanoid models controlled by a set of pre-specified parameters such as limb lengths that are used to vary shape [12, 13, 16]; 4) data driven models where human body shape variation is learned fr...

Citation Context

...-squares techniques [10]. 3 Models In this section, we introduce the models we use to describe both body pose and shape at a given time as well as its motion over time. 3.1 Body Model In earlier work =-=[9]-=-, we have developed a body-modeling framework that relies on attaching implicit surfaces, also known as soft objects, to an articulated skeleton. Each primitive defines a field function and the skin i...

Citation Context

..., (12) where the θ k j represents the vector of individual joint angles at phase µk, defined as the j’th component of ψ(µk, α1, · · · , αm) in Eq. 6. The derivatives of Ft with can be easily computed =-=[45]-=-. Because respect to the D individual joints angles ∂Fo,t ∂θ k j the θk j are linear combinations of the Θk ij eigen-poses, ∂θk j ∂αi coordinate of Θk i . Similarly, we can write reduces to Θk ij, the...

Citation Context

... where the motion of points located on the skin is computed using so-called skinning techniques from Computer Animation, and the skin surface position, used for EM-ICPS, is defined using Soft Objects =-=[13]. -=-The rest of the paper is organized as follows: – Section 2 presents the framework of tracking a rigid body of known shape (an ellipsoid in this case) from noisy 3D points trajectories using the EM-I...

Citation Context

...urier analysis by the fitting of our PCA-based motion models. 3 Tracking In this section we introduce our body and motion models and show how we can use them for tracking. 3.1 Models In previous work =-=[18]-=- we have developed the body-model depicted by Figure 1(a) that is composed of implicit surfaces attached to an articulated skeleton. Each primitive defines a field function and the skin 2s1 0.9 0.8 0....

Citation Context

...TTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 33, NO. 3, MARCH 2011 full advantage of the rigid point registration algorithm, which is quite different from data-point-toobject-part registration [34], =-=[35]-=-, [36], [37], [38]. We note that our method is similar in spirit to [39]. However, the latter suffers from the limitations of ICP. . One important property of any point registration method is its robu...

Citation Context

... solutions using a single video camera have been explored [6, 22]. “However, they are not always robust, in part because image data is inherently noisy and in part because it is inherently ambiguous” =-=[15, 19]-=-. We believe the use of depth information can significantly reduce the ambiguity and make one-camera solutions practical. In this regard, our approach is mostly related to methods that use stereovisio...