A real-time method for deriving accurate body-part labels and basic human posture from monocular, active camera image sequences is presented. Bootstrapping parameters, such as human clothing models or a background model of the environment are not required. The ultimate goal of the research is to track human gestures using active PTZ cameras and hence the human segmentation must operate in real-time. Since such cameras are free to pan, tilt and zoom, a background model of the scene is not available. By fusing the outputs from various temporal filters and a statistical skin colour model, and then ap-plying novel spatial and temporal skin linkage stages, the system is capable of reliably associating and labelling hands and heads, as well as providing clues as to the basic forms of multiple people within its field of view after just a few seconds of observations.
Real-Time Skin Labelling in Active Camera Images
Chippendale, Paul Ian
2005-01-01
Abstract
A real-time method for deriving accurate body-part labels and basic human posture from monocular, active camera image sequences is presented. Bootstrapping parameters, such as human clothing models or a background model of the environment are not required. The ultimate goal of the research is to track human gestures using active PTZ cameras and hence the human segmentation must operate in real-time. Since such cameras are free to pan, tilt and zoom, a background model of the scene is not available. By fusing the outputs from various temporal filters and a statistical skin colour model, and then ap-plying novel spatial and temporal skin linkage stages, the system is capable of reliably associating and labelling hands and heads, as well as providing clues as to the basic forms of multiple people within its field of view after just a few seconds of observations.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
