Extracting low-level and mid-level descriptors from a soccer match to generate a summary of soccer highlights.

Automatic generation of sports highlights from recorded audiovisual content has been object of great interest in recent years. The problem is indeed especially important in the production of second and third division highlights videos where the quantity of raw material is significant and does not contain manual annotations. In this thesis, a new approach for automatic generation of soccer highlights is proposed. The approach is based on the segmentation of the video sequence into shots that will be further ana- lyzed to determine its relevance and interest. For every video shot a set of low and mid level audio-visual descriptors are computed and combined in order to obtain different relevance measures based on empirical knowledge rules. The final summary is generated by selecting those shots with highest interest according to the specifications of the user and the results of relevance measures. The main novelties of this work have been the temporal combination of two shot boundary detectors; the selection of keyframes using motion and color features; the generation of new soccer audio mid-level descriptors; the robust detection of soccer players; the employment of a novel object detection technique to spot goal-posts and finally, the creation of a flexible and user-friendly highlight gen- eration framework. The thesis is mainly devoted to the description of the global visual segmentation module, the selection of audiovisual descriptors and the general scheme for evaluating the measures of relevance. Several results have been produced using real soccer video sequences that prove the validity of the proposed framework.