Design and Implementation of a Content-Based Video Retrieval System Using Moving Objects Trajectories

Abstract

In this paper, we design and implement a content-based video retrieval system using moving objects’ trajectories. For this, we first present a spatio-temporal representation scheme which can efficiently deal with a single moving object as well as multiple moving objects. Secondly, we describe an efficient similar sub-trajectory retrieval scheme based on a new k-warping distance algorithm. Thirdly, we propose a new signature-based access method which can deal with a large amount of database and can support efficient retrieval on a user query based on moving objects’ trajectories. To show the usefulness of applying our above schemes to a real video database application, we implement a content-based soccer video retrieval system, called CSVR, because the soccer video data have a lot of motion information based on a soccer ball being a salient object. The CSVR provides graphic user interfaces for both soccer video indexing and retrieval. Finally, we do the performance analysis of our similar sub-trajectory retrieval scheme based on our k-warping distance algorithm.