Contour Tracking By Stochastic Propagation of Conditional Density：（轮廓跟踪的随机传播条件密度）.pdfVIP

In Pro c Europ ean Conf Computer Vision pp Cambridge UK Contour tracking by sto chastic propagation of conditional density Michael Isard and Andrew Blake Department of Engineering Science University of Oxford Oxford OX PJ UK fmisard ab grob otsoxacuk Abstract The problem of tracking curves in dense visual clutter is a challenging one Trackers based on Kalman lters are of limited use b ecause they are based on Gaussian densities which are unimo dal they cannot represent simultaneous alternative hyp otheses Extensions to the Kalman lter to handle multiple data asso ciations work satisfactorily in the simple case of p oint targets but do not extend naturally to con tinuous curves A new sto chastic algorithm is prop osed here the Con densation algorithm Conditional Density Propagation over time It uses factored sampling a metho d previously applied to interpretation of static images in which the distribution of p ossible interpretations is represented by a randomly generated set of representatives The Con densation algorithm combines factored sampling with learned dynam ical mo dels to propagate an entire probability distribution for ob ject p osition and shap e over time The result is highly robust tracking of agile motion in clutter markedly sup erior to what has previously b een attainable from Kalman ltering Notwithstanding the use of sto chastic metho ds the algorithm runs in near realtime The problem of tracking curves in clutter The purp ose of this pap er is to establish a sto chastic framework for tra