Markus Middledorf and Hans-Helmut Nagel
Institute of Algorithms and Cognitive Systems, University of Karlsruhe, Germany.
In previously published approaches (ECCV '98, DAGM '98), we have investigated estimation and segmentation of Optical Flow (OF) fields in monocular image sequences recorded from road intersection scenes, as well as tracking of OF segments without assumptions about the shape of vehicles or the lane structure of an intersection. These approaches had been improved in the meantime in order to create more robust versions. In this contribution, we show that the system version available now allows us to track vehicles, using essentially the same parameters as before, in the monocular image sequence made available for this workshop -- despite the fact that more than one condition implied by the image sequences previously used by us are no longer satisfied. We report the steps taken in order to cope with the new sequence and the encouraging results obtained thereby. A number of additional experiments are stimulated by these new experiences which are discussed in the concluding section of this contribution.