A Synthetic Video Dataset Generation Toolbox for Surveillance Video Synopsis Applications

Video synopsis technique aims to condense long duration videos into its compact representation for efficient browsing and retrieval of surveillance videos. The synopsis videos depend on the results of object detection and tracking in input video. However, there is a lack of publicly available commonly used datasets that are properly annotated for training trackers and analyzing the performance of various approaches in video synopsis. This paper introduces an interactive toolbox that allows users to generate synthetic videos with user-defined parameters and related tube information, eliminating the steps of detection and tracking. The proposed toolbox enables users to simulate general and specific scenarios of interest, which are expected to be observed in surveillance videos. We present experiments that show the usability and effectiveness of using this toolbox in evaluating different video synopsis methods.