Smart particle filtering for 3D hand tracking

Bray, Matthieu; Koller-Meier, Esther; Gool, Luc Van

doi:10.1109/afgr.2004.1301612

Cited by 86 publications

(85 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The skin surface has been modeled as a triangulated surface in [2], a simplex mesh in [7] while small set of simple primitives (conics and convex polyhedron) have been adopted in [4,18,19]. We choose the latter for the main reason that the use of mesh would leads to undifferentiable silhouette position with respect to parameters θ and would hinder the local search (see [8]).…”

Section: Hand Silhouette Computationmentioning

confidence: 99%

“…To evaluate the likelihood of a plausible candidate 3D configuration, we synthesize the corresponding hand silhouette projections in the image plane and measure their likelihood given a generative model for the background and hand skin pixels. In order to avoid convergence in local minima, we combine our approach with the smart-particle filter [2]. The latter method exploits depth information , an important limitation that is not present in our method.…”

Section: Introductionmentioning

confidence: 99%

“…Model-driven algorithms are the most common selection of such methods. The hand model is then matched to the 3D disparity map obtained from the stereo-matching algorithm [2,3]. However, one can question the accuracy of such disparities estimation due to the absence of strong features on a hand.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Monocular Hand Pose Estimation Using Variable Metric Gradient-Descent

Gorce¹,

Paragios²

2006

Procedings of the British Machine Vision Conference 2006

View full text Add to dashboard Cite

In this paper, we propose a novel model-based approach to recover 3D hand pose from 2D images through a compact articulated 3D hand model whose parameters are inferred in a Bayesian manner. To this end, we propose generative models for hand and background pixels leading to a loglikelihood objective function which aims at enclosing hand-like pixels within the silhouette of the projected 3D model while excluding background-like pixels.Segmentation and hand pose estimation are unified through the minimization of a single likelihood function, which is novel and improve overall robustness. We derive the gradient in the hand parameter space of such an area-based objective function, which is new and allows faster convergence rate than gradient free methods. Furthermore , we propose a new constrained variable metric gradient descent to speed up convergence and finally the so called smart particle filter is used to improve robustness through multiple hypotheses and to exploit temporal coherence. Very promising experimental results demonstrate the potentials of our approach.

show abstract

Section: Hand Silhouette Computationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Monocular Hand Pose Estimation Using Variable Metric Gradient-Descent

Gorce¹,

Paragios²

2006

Procedings of the British Machine Vision Conference 2006

View full text Add to dashboard Cite

show abstract

“…In [4], structured light was used to acquire 3D depth data; however, skin color was used for segmenting the hand as well, which requires homogeneous and interest points on the surface of the hand using a stereo camera. Motion information obtained from the 3D trajectories of the points was used to augment the range data.…”

Section: Related Workmentioning

confidence: 99%

Gesture Recognition Based Mouse Events

Puri¹

2013

IJCSIT

View full text Add to dashboard Cite

show abstract

“…It approximates the function of probability density through finding a group of random sample in state space and takes the mean value replacing the integration calculation. The tracking algorithm based on Particle Filter can maintain multiple assumptions and solve some non-linear problems [8,9], especially handle the background interfered and restore from wrong tracking. Usually it can get stable result [10,11].…”

mentioning

confidence: 99%