Tracking of motor vehicles from aerial video imagery using the OT-MACH correlation filter

Accurately tracking moving targets in a complex scene involving moving cameras, occlusions and targets embedded in noise is a very active research area in computer vision. In this paper, an optimal trade-off maximum correlation height (OT-MACH) filter has been designed and implemented as a robust tracker. The algorithm allows selection of different objects as a target, based on the operator’s requirements. The user interface is designed so as to allow the selection of a different target for tracking at any time. The filter is updated, at a frequency selected by the user, which makes the filter more resistant to progressive changes in the object’s orientation and scale. The tracker has been tested on both colour visible band as well as infra-red band video sequences acquired from the air by the Sussex County police helicopter. Initial testing has demonstrated the ability of the filter to maintain a stable track on vehicles despite changes of scale, orientation and lighting and the ability to re-acquire the track after short losses due to the vehicle passing behind occlusions.