PTZ Handover: Tracking an object across multiple surveillance cameras

Sammanfattning: Tracking objects in a scene is a crucial task in accomplishing surveillance that enhances security and provides valuable information about the events happening at the site. For this task, the PTZ (pan-tilt-zoom) cameras can be utilized to achieve fluid tracking as they provide all-around surveillance with zoom capabilities. The drawback of current tracking solutions is the lack of interoperability between cameras, e.g. to signal the position of an object so that multiple cameras can track it simultaneously. This project highlights the importance of continuously tracking an object across a site and proposes a solution on how to handover the target from one camera to another. Thus the need of performing PTZ coordinate transformation is necessary to direct multiple PTZ cameras toward the same target. For simplicity, the scope of the project was limited to a system consisting of only two cameras, with a focus on tracking one object at a time. The method consists of two steps: namely to perform a calibration procedure to determine the spatial relationship between two cameras and to then track a single object across a site. The tracking process is handled by a centralized server, which determines which objects to track, where to position the cameras and when to perform the handover. The results show that tracking an object across two cameras, mounted at different heights and located multiple meters apart, is fully achievable. Even though the built-in tracker can be perceived as slightly delayed, the handover functionality still managed to execute as expected even with the target moving at a moderately high velocity. The output of the calibration was found to be rather satisfactory, but could however be refined to achieve even higher accuracy. In conclusion, the proposed solution works well and entails that this kind of functionality may further enhance all-around surveillance. As future work, the calibration procedure can easily be expanded to multiple cameras, but tracking multiple objects at the same time requires advanced theoretical investigation.