The cat solution is based on an iterative optimization process,  with the aim to find a consesus between calibration, human detections, tracking and trajectory associations. The project deals with slightly overlapping as well as the challenging case of non-overlapping camera views, where there are no point correspondences across the views. Instead human detections, trajectories, human velocity smoothness are the cues which are of use in such scenarios. An approach pursuing these goals has a potential to become a major technology for networked surveillance cameras.

The particular challanges of the project are

• to achieve a reliable approach for practical use by non-expert

• to minimize human interaction

• to use tracked humans as calibration targets while dealing with typical inaccuracies of the human detectors

• achieve accuracy and robustness enough for better re-detection of people, their tracking and trajectory associations

• achieve run-times within some minutes on an off-the-shelf computer