Center for Robust Speech Systems


Automatic Highlights Generation using an Information-theoretic excitability measure

This page contains some video samples of automatically generated highlights of Baseball games using the methods presented in our paper:

  • Taufiq Hasan, Hynek Boril, Abhijeet Sangwan and John H. L. Hansen,"Multi-modal highlight generation for sports videos using an information-theoretic excitability measure", EURASIP Journal on Advances in Signal Processing, Nov. 2013. [pdf]

  • First video shows the extracted highlight video from a full Baseball game where in total 206 pitching event takes place. It is reduced down to a duration of 2:12 using the highlights generation algorithm. In this processing the pitching times are assumed to be known (used the manual labels) but all the features (slow motion, excitement in audio, scene cut and video motion) are extracted automatically.

    Second video shows the highlights of the same game using the automatic pitching scene detection for segmentation. Due to errors in the detection some exciting events may not have been included in the highlight. This can be further improved by a more sophisticated pitching time detection, which was outside the scope of the current work. The Sample Highlights 1 video represents most of the exciting events in the game demonstrating the effectiveness of the proposed excitability measure.