MODELING MOVEMENT RELATIONS IN DYNAMIC URBAN SCENES
Panagiotis Partsinevelos, Peggy Agouris, Anthony Stefanidis, Panayotis Partsinevelos · 2000
In this paper we address the problem of analyzing and managing movement in dynamic scenes captured in video datasets. We present an approach to summarize the important information included in video datasets by analyzing the trajectories of objects within them. Trajectories form the basis for the construction of abstract data types upon which similarity of behaviors and expected behavior processes take place. We base our selection of representative time instances on the segmentation of trajectory lines into break points termed “nodes”. The nodes are distributed dynamically to capture the information content of regions within the 3-D spatiotemporal space. They are computed through self-organizing maps of neural network processing. Additional nodes are supplementing the procedure and are originated from reasoning and proximity analysis. Topologic relations between moving objects in the scene and dynamic topology of the trajectories are processed in order to include the significant information of movement relations in the product summary. This work provides a novel approach to manage dynamic scene analysis at higher levels of