A Novel Algorithm of Tube Association and Rearrangement in Multi-camera Video Synopsis

To solve the problems of storage and browsing through massive surveillance videos in the current society, the video synopsis was studied in depth. There are few optimal solutions to solve the following two problems of multi-camera video synopsis: 1) The association and stitch of image sequences (also known as tube) of the same object across space and time under different cameras. 2) Rearrangement of global tubes in the global context. In this paper, we propose a novel video synopsis algorithm for partially overlapping cameras. Our main contributions reside in two aspects: Firstly, our algorithm can accurately associate tubes across multi-camera through a hierarchical clustering algorithm based on an adaptive threshold. It is more suitable for tubes that are broken and objects which pace back and forth. Secondly, our algorithm uses a novel priority-based tube rearrangement to maintain the integrity of the global tubes activity of the synopsis video. We have demonstrated the effectiveness of our approach from the Toulouse campus surveillance dataset and another surveillance dataset captured by our camera network.