We present an approach for recovering articulated body pose fromsingle monocular images using the Specialized Mappings Architecture(SMA), a nonlinear supervised learning architecture. SMAs consist ofseveral specialized forward (input to output space) mapping functionsand a feedback matching function, estimated automatically from data.Each of these forward functions maps certain areas (possiblydisconnected) of the input space onto the output space. A probabilisticmodel for the architecture is first formalized along with a mechanismfor learning its parameters. The learning problem is approached using amaximum likelihood estimation framework; we present expectationmaximization (EM) algorithms for several different choices of thelikelihood function. The performance of the presented solutions underthese different likelihood functions is compared in the task ofestimating human body posture from low-level visual features obtainedfrom a single image, showing promising results
展开▼
