Research On Video Abstraction Technology For Basketball Game

Video abstraction technology has always been a hot topic in the field of multimedia. Due to the particularity of basketball video, traditional video abstraction techniques cannot effectively extract the key information. Video abstraction for basketball requires domain knowledge to extract the highlights, which has high practical value and broad application scenario.In this paper, we review the research progress of basketball video abstraction techniques, current problems including: low accuracy of highlight extraction and poor versatility, lack of effective highlight sorting model, etc. To solve these problems, we propose an improved method.We present an efficient score board detection algorithm. Both teams’score on the score board is recognized using classification method. Scoring event is detected when score changes. Boundary of each scoring event can be found by shot segmentation. Finally we extract all scoring shots and synthesis them into a video skim.In this paper, we propose a brand new highlight sorting model. We describe scoring shot’s exciting degree by both scoring features (offense time, buzzer beater, and-one, etc.) and audio features (short-time energy, zero-crossing rate). Learning to rank approach is used in our system for ranking multi-feature data. Then top five scoring shots are extracted into top five highlights.We implement a basketball video abstraction system. For each input basketball video, we extract two video skims:scoring shot highlights and top five highlights. We get over 93% accuracy and recall on scoring shot extraction through experiment. Top five highlights also achieve reasonable result. Experimental results show that our method is feasible and effective. This method can also be extended to other score-driven sports video abstraction.