This paper presents a particle filteringframework for tracking multiple persons with amonocular camera. So far, most of techniques basedon particle filtering rely on an assumption thatmeasurements on pose and speed of moving personsare sufficiently precise. Unfortunately, such anassumption is often violated due to measurementnoises as well as irregular movements of persons. Todeal with the problem, we have developed atechnique for measuring relative position betweenpersons using occlusion reasoning. In particular, weshow how the use of relative position measurementscan improve the tracking performance, even indifficult situations where two persons frequentlyoverlap in images.
展开▼
