A Modeling and Similarity Measure Function for Multiple Trajectories in Moving Databases

Abstract

In this paper, we focus on a new spatio-temporal representation scheme which can efficiently model multiple trajectories based on several moving objects in video databases. The traditional methods only consider direction property, time interval property, and spatial relations property for modeling moving objects’ trajectories. But, our method also takes into account on distance property, conceptual location information, and related object information (e.g. player name having a soccer ball). In addition, we propose a similarity measure function that improves a retrieval accuracy to measure a similarity among multiple trajectories. The proposed scheme supports content-based retrieval using moving objects’ trajectories and supports semantics-based retrieval using concepts which are acquired through the location information of moving objects. Finally, from the experimental results using real trajectories extracted from soccer video data with soccer ball and player, the performance of our scheme achieves about 15-20% performance improvement against existing schemes when the weights of angle and topological relation are over two times than that of distance.