Computer Vision News - April 2016

Application Video Synopsis by BriefCam COMPUTER VISION NEWS One of the main problems in video search by humans is their length: if we want to observe events in a video of several hours, like what happened in a certain place during a daytime, the result may be a new video of one or two hours. That, even after removing eventless sections! Of course, it is computer vision technology that provides the ideal solution, but it is interesting to learn the conceptual revolution that was necessary to make it happen. The name behind this innovation is BriefCam , who provides an application able to summarize hours of video into minutes. We asked the co-founder of BriefCam, Professor Shmuel Peleg of the Hebrew University of Jerusalem , to tell us how he brought about this Video Synopsis solution. Genesis of a solution According to Prof. Peleg, algorithms can detect people and objects, once they have been trained in advance to search for that particular category of items. But since computers are not ready yet to understand the way humans do, they cannot decide which parts of a video are significant for us. The only solution is therefore to break the conceptual problem laying in the chronology of video: as long as we look at videos as a sequence of chronological frames which cannot be altered, we cannot generate shorter summaries. Video Synopsis allows people to be featured together in the video, though they were never in the same place at the same time: two cars passing at 8 AM and 4 PM are shown together, in a joint vision of events that happened at different times. Prof. Peleg says that going beyond chronology breaks the biggest of video taboos . But the concept of video frame has to be eradicated, in order to create efficient video summaries. 10