MATCHING LOGOS FOR SLOW MOTION REPLAY DETECTION IN BROADCAST SPORTS VIDEO

Slow motion replays are usually linked with semantically important highlights in broadcast sports video. A replay often happens between two logo transition sequences, which can be used to locate replays. In this paper, we present a new method to extract replays based on automatic logo template generation and logo searching. The method utilizes speeded-up robust features (SURF) to find repeating logo patterns and then to search those patterns in the video. Compared with the existed replay detection methods based on finding logo patterns, our method can handle more complex logo transition types. It can also be applied to the cases where different logo transitions are used in one match. We build a dataset consisting of 28 different logo transition types and 6 sport genres with 3907 minutes in total. Good experimental results have been achieved on the dataset and it validates the effectiveness and robustness of the proposed method.