Hybrid multi scale hard switch YOLOv4 network for cricket video summarization

Article; Early Access

关键词: SPORTS VIDEO;    ATTENTION;    REPLAY;   

DOI  :  10.1007/s11276-023-03449-8

来源: SCIE

【 摘 要 】

Cricket is a popular sport with a lengthy duration that makes it challenging to watch in its entirety. Therefore, video summarization techniques are essential to providing viewers with a condensed version of the match's exciting moments. Automated cricket video summarization is difficult due to the sport's regulations and extended sessions. Existing methods often include repetitive shots, making the summary less concise and informative. Therefore, this paper proposes a hybrid video summarization framework that uses audio and text features to extract exciting clips from the raw cricket video. The framework employs the Multi-Scale Hard Switch YOLOv4 (MSHS-YOLOv4) network to accurately detect and label exciting events, including small details such as a ball hitting the stumps. A significance score is computed for each event to generate a summary that includes the most exciting and significant moments. The proposed method eliminates replay shots, reducing redundancy and making the summary more concise. The proposed method combines audio and video features to identify the most exciting moments, uses the MSHS-YOLOv4 network to detect and label exciting events, computes a significance score for each event, and eliminates replay shots to generate a concise summary. The proposed method outperforms existing summarization techniques in terms of accuracy, precision, recall, F1-score, and error. The analysis shows a significant increase in performance compared to the existing methods.

【 授权许可】

Free