My research group, previously called the “Visual Surveillance Group”, has emphasized the visual aspects of detecting, describing, identifying and analyzing objects, people, activities and events in both images and videos. We are continuing along this path, except for three aspects:

  1. Our primary focus is now on ice hockey game and based on television broadcast videos. Thus, the objects involved are players on skates, pucks, hockey sticks, nets, etc. We seek to employ the video to provide a detailed and analytical description of a game.
  2. The broadcast video provides a two-dimensional picture or set of frames in sequence as a function of time. The video also contains continuous simultaneous descriptions of these frames in both audio and closed captions (language describing the play-by-play). We refer to these three sets of data as modes, and to the research activity, as the study of Intelligent Multimodal Video Content.
  3. Machine Learning is being employed to teach Neural Networks to achieve the above analyses. This hockey study uses deep learning much more prominently than our earlier work because of the nature and complexity of the input data.

In order to perform the video analysis of a hockey game, we assume the existence of a single broadcast video, which in reality is compiled by using multiple video cameras placed around a hockey rink. Camera operators are able to manually rotate and translate a camera in 3D-space, as well as adjust the focus. Thus, an on-air game video is actually a heavily edited version of many sequences of close-up, mid-range, and long-range views. Unfortunately, these multiple videos are not publicly available! Only the so-called broadcast video, which is an amalgamation of all of the on-air video clips shown on TV, is circulated to the public.

Therefore, we base our research on these broadcast videos and develop methodologies to address the inherent challenges caused by the nature and complexity of these videos. For more information, please see the Research section.

Copyright © 2019 Intelligent Multimodal Video Analysis Group