Date of Award


Publication Type

Master Thesis

Degree Name



Computer Science

First Advisor

Boufama, Boubakeur (School of Computer Science)


Artificial Intelligence.




Disjoint intra-camera tracking is the task of tracking objects across video-surveillance cameras that have non-overlapping views. Disjoint intra-camera tracking is difficult due to the gaps in observation as an object moves between camera views. To solve this problem, an intra-camera video-surveillance system builds an appearance profile of the objects seen in its camera, and matches these appearance profiles to achieve the effect of tracking. This thesis demonstrates two novel ideas that improve Disjoint intra-camera tracking. The first is to use a Zernike moment based shape feature for objects observed in a scene, used to describe the shape of an object in a compact, reliable form. The second is to dynamically weigh the Zernike moment shape feature with other standard features to achieve better tracking results. Weighting emphasis is given to better features, more stable features, more recent values of features, and features that have been reliably translated from a different video camera.