Video Browsing Using Object Trajectories
Publication from Digital
Lee F., Werner Bailer
Taipei, Taiwan Proceedings of 17th Multimedia Modeling Conference, Volume 1 / 1973 - Volume 6609 / 2011, 1/2011
Video browsing methods are complementary to search and retrieval approaches, as they allow for exploration of unknown content sets. Objects and their motion convey important semantics of video content, which is relevant information for video browsing. We propose extending an existing video browsing tool in order to support clustering of objects with similar motion and visualization of the objects positions and trajectories. This requires the automatic extraction of moving objects and estimation of their trajectories, as well as the ability to group objects with similar trajectories. For the first issue we describe the application of a recently proposed motion trajectory clustering algorithm, for the second we use k-medoids clustering and the dynamic time warping distance. We present evaluation results of both steps on real world traffic sequences from the Hopkins155 data set. Finally we describe the description of analysis results using MPEG-7 and the integration into the video browsing tool.