Density Based Modeling 3 D Shape Estimation Based

Density Based Modeling 3 D Shape Estimation Based on Density Driven Model Fitting 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Introduction Objectives • locate body and body parts, • estimate their shapes, • track individual body parts, and • recognize and interpret gestures. Need a shape model • flexible description of elaborate shapes • efficient & robust fitting procedure 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Density fields? Use parametric density field for • target shape family representation, and • as model fitting driving force 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Density fields! (consistency) (conservation) 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Optimal model fit maximizeθ F(θ) subject to some constraints on θ 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Conservation and consistency vs scale 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Relaxing conditions 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling articulated objects (cumulative density) (component density) 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Density based model of a human body head: torso: limbs: 0. 5 -3 -2 -1 0 1 2 x 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA 3

Density Based Modeling Model fitting concerns and details but 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Hierarchical fitting 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Experiments with synthetic data 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Experiments with real data 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Density Based Modeling Summary Major contributions • generic density-based shape modeling framework • hierarchical model fitting • shape modeling to any precision Differences from similar approaches • models with low number of degrees of freedom • works with volumetric shape data • generalizes to any number of dimensions Applications • recognition and tracking of non-rigid articulated objects • physically realistic modeling and visualization • flexible and efficient shape representation 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA

Prospective Research Immediate goals • more precise 3 D shape recognition, • movement analysis of articulated objects Longer term goals • tracking multiple people and objects • 3 D monitoring of complex activities • automatic shape discovery Far reaching plans • custom augmented worlds, virtual reality games • vision-based human-computer interfaces 2/13/2022 E. Borovikov and L. Davis, University of Maryland, USA