This paper looks into the problem of pedestrian tracking using a monocular,potentially moving, uncalibrated camera. The pedestrians are located in eachframe using a standard human detector, which are then tracked in subsequentframes. This is a challenging problem as one has to deal with complexsituations like changing background, partial or full occlusion and cameramotion. In order to carry out successful tracking, it is necessary to resolveassociations between the detected windows in the current frame with thoseobtained from the previous frame. Compared to methods that use temporal windowsincorporating past as well as future information, we attempt to make decisionon a frame-by-frame basis. An occlusion reasoning scheme is proposed to resolvethe association problem between a pair of consecutive frames by using anaffinity matrix that defines the closeness between a pair of windows and then,uses a binary integer programming to obtain unique association between them. Asecond stage of verification based on SURF matching is used to deal with thosecases where the above optimization scheme might yield wrong associations. Theefficacy of the approach is demonstrated through experiments on severalstandard pedestrian datasets.
展开▼
