This work presents an adaptive video fast-forwarding technique that helps users browse long and untrimmed videos quickly. The proposed method is built upon the concept of the elastic timeline that a) allows users to select low-level cues of a video that they like to enhance and b) adaptively fast-forwards the video while keeping scenes with the selected cues played at a lower speed. The proposed method helps users to choose appropriate cues, in particular, a small set of object cues arranged automatically tailored to the contents of a given video out of many. Our user studies show that object cues selected by our method allow users to browse videos recorded under a variety of scenarios, such as cooking at home, strolling in the street, and playing in an amusement park, more quickly compared to the state-of-the-art interface with the elastic timeline in event finding tasks.

This work was supported by JST ACCEL (grant JPMJAC1602) and JST CREST (grant JPMJCR14E1).