Learning and Inference in Parametric Switching Linear Dynamic SystemsReportar como inadecuado


Learning and Inference in Parametric Switching Linear Dynamic Systems


Learning and Inference in Parametric Switching Linear Dynamic Systems - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

We introduce parametric switching linear dynamicsystems P-SLDS for learning and interpretation ofparametrized motion, i.e., motion that exhibits systematictemporal and spatial variations. Our motivating exampleis the honeybee dance: bees communicate the orientationand distance to food sources through the dance angles andwaggle lengths of their stylized dances. Switching linear dynamicsystems SLDS are a compelling way to model suchcomplex motions. However, SLDS does not provide a meansto quantify systematic variations in the motion. Previously,Wilson and Bobick presented parametric HMMs 21, an extensionto HMMs with which they successfully interpretedhuman gestures. Inspired by their work, we similarly extendthe standard SLDS model to obtain parametric SLDS.We introduce additional global parameters that representsystematic variations in the motion, and present generalexpectation-maximization EM methods for learning andinference. In the learning phase, P-SLDS learns canonicalSLDS model from data. In the inference phase, P-SLDSsimultaneously quantifies the global parameters and labelsthe data. We apply these methods to the automatic interpretationof honey-bee dances, and present both qualitativeand quantitative experimental results on actual bee-trackscollected from noisy video data.



Computational Perception and Robotics - Computational Perception and Robotics Publications -



Autor: Oh, Sang Min - Rehg, James M. - Balch, Tucker - Dellaert, Frank - -

Fuente: https://smartech.gatech.edu/







Documentos relacionados