This paper proposes a framework for modeling interactions in muliple object 3D Bayesian tracking. It exploits both the computational cheapness of independent single object filters and the modeling power of the joint filter. Sampling from complex joint propagation densities is avoided by introducing interaction a posteriori, after blind single object propagation has been performed. To deal with occlusions for each object a support layer is computed. It contains probabilistic information about how likely a pixel is occluded by another object. Utilized to give less weight to likely occluded pixels, it provides the basis of a robust likelihood model. The implementation of the proposed ideas in a Sequential Monte Carlo framework are discussed. Experiments on a synthetic data show the robustness of the proposed ideas
Modeling Interactions in Multiple Object Bayesian Tracking
Lanz, Oswald
2004-01-01
Abstract
This paper proposes a framework for modeling interactions in muliple object 3D Bayesian tracking. It exploits both the computational cheapness of independent single object filters and the modeling power of the joint filter. Sampling from complex joint propagation densities is avoided by introducing interaction a posteriori, after blind single object propagation has been performed. To deal with occlusions for each object a support layer is computed. It contains probabilistic information about how likely a pixel is occluded by another object. Utilized to give less weight to likely occluded pixels, it provides the basis of a robust likelihood model. The implementation of the proposed ideas in a Sequential Monte Carlo framework are discussed. Experiments on a synthetic data show the robustness of the proposed ideasI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
