Temporal video analysis
Dong‐Jun Park, David A. Eichmann · 2008
Detection of events and actions in video entails substantial processing of very large, even open-ended, video streams. The video information retrieval community has substantially focused on a single frame approach for retrieval and classification tasks. While this approach is sufficiently powerful for certain types of semantic concepts, there are more complicated categories such as events or motion that require more than that provided by a single frame. We present a simple and effective way of extracting time variant information in video data using three perspective views into a video stack.