Robust facial feature tracking under varying face pose and facial expression
Pattern Recognition
| Hi-index | 0.00 |
This paper proposes a switching hypothesized measurements (SHM)model supporting multimodal probability distributions and presentsthe application ofthe model in handling potential variability invisual environments when tracking multiple objects jointly. For aset of occlusion hypotheses, a frame is measured once under eachhypothesis, resulting in a set of measurements at each timeinstant. A computationally efficient SHM filter is derived foronline joint region tracking. Both occlusion relationships andstates of the objects are recursively estimated from the history ofhypothesized measurements. The reference image is updatedadaptively to deal with appearance changes of the objects. The SHMmodel is generally applicable to various dynamic processes withmultiple alternative measurement methods.