Video condensation for video forensics

Video forensics has become an important issue because of the setup of a large number of surveillance cameras. Video condensation can increase efficiency in quickly searching criminal events from the vast amounts of surveillance video data. In this paper, we propose a video condensation technique for video forensics to condense a long period of surveillance video into a short video clip under keeping important information. The proposed technique is comprised of the following three steps: image sequences transformation, motion oriented filter, and ribbon carving. First, the original video sequence is converted into X-T and Y-T image sequences. Then, a motion oriented filter operator based on Gaussian gradient process is applied to extract moving-object maps in a specific direction from the X-T and Y-T image sequences. Finally, we apply a ribbon carving approach to achieve the purpose of video condensation. Preliminary experiments on realistic video data demonstrate the applicability of the proposed technique.