Switching Linear Dynamic System (SLDS) models are a popular technique for modeling complex nonlinear dynamic systems. An SLDS can describe complex temporal patterns more concisely and accurately than an HMM by using continuous hidden states. However, the use of SLDS models in practical applications is challenging for three reasons. First, exact inference in SLDS models is computationally intractable. Second, the geometric duration model induced in standard SLDSs limits their representational power. Third, standard SLDSs do not provide a principled way to interpret systematic variations governed by higher order parameters.The contributions in this paper address all of these three challenges. First, we present a data-driven MCMC (DD-MCMC) sampling method for approximate inference in SLDSs. We show DD-MCMC provides an efficient method for estimation and learning in SLDS models. Second, we present segmental SLDSs (S-SLDS), where the geometric distributions of the switching state durations are replaced with arbitrary duration models. Third, we extend the standard SLDS model with additional global parameters that can capture systematic temporal and spatial variations. The resulting parametric SLDS model (P-SLDS) uses EM to ro-S.M. Oh ( ) · J.M. Rehg · T. Balch · F. Dellaert bustly interpret parametrized motions by incorporating additional global parameters that underly systematic variations of the overall motion.The overall development of the extensions for SLDSs provide a principled framework to interpret complex motions. The framework is applied to the honey bee dance interpretation task in the context of the on-going BioTracking project at the Georgia Institute of Technology. The experimental results suggest that the enhanced models provide an effective framework for a wide range of motion analysis applications.

No abstract

Abstract-We aim to perform robust and fast vision-based localization using a pre-existing large map of the scene. A key step in localization is associating the features extracted from the image with the map elements at the current location. Although the problem of data association has greatly benefited from recent advances in appearance-based matching methods, less attention has been paid to the effective use of the geometric relations between the 3D map and the camera in the matching process.In this paper we propose to exploit the geometric relationship between the 3D map and the camera pose to determine the visibility of the features. In our approach, we model the visibility of every map feature with respect to the camera pose using a nonparametric distribution model. We learn these nonparametric distributions during the 3D reconstruction process, and develop efficient algorithms to predict the visibility of features during localization. With this approach, the matching process only uses those map features with the highest visibility score, yielding a much faster algorithm and superior localization results. We demonstrate an integrated system based on the proposed idea and highlight its potential benefits for the localization in large and cluttered environments.

No abstract

We introduce mixture trees, a tree-based data-structure for modeling joint probability densities using a greedy hierarchical density estimation scheme. We show that the mixture tree models data efficiently at multiple resolutions, and present fast conditional sampling as one of many possible applications. In particular, the development of this datastructure was spurred by a multi-target tracking application, where memory-based motion modeling calls for fast conditional sampling from large empirical densities. However, it is also suited to applications such as texture synthesis, where conditional densities play a central role. Results will be presented for both these applications.