Multi-modal, Multi-labeled Sports Highlight Extraction
| Autor: | Hsing-Kuo Pao, Jiabin He |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Focus (computing)
Modalities Computer science business.industry Feature extraction Frame (networking) 020206 networking & telecommunications 02 engineering and technology Machine learning computer.software_genre Variety (cybernetics) Task (project management) Modal 0202 electrical engineering electronic engineering information engineering Task analysis 020201 artificial intelligence & image processing Artificial intelligence business computer |
| Zdroj: | TAAI |
| DOI: | 10.1109/taai51410.2020.00041 |
| Popis: | Video highlight extraction has been as important as many other research on video analysis. To focus on sports highlight extraction, we aim to adopt machine learning techniques to provide an efficient approach to catch the most important moments in sports games of only a short period of time in such a fast-forward age. To elaborate the procedure, we examine frame by frame in a video to summarize what happened in a sports game which may contain the exciting moments, funny moments, game turning point, or even controversial events to name a few.One of the special treatment in this work is that we emphasize the multi-modal approach. Sports video contains a variety of modalities such as images, audios, scores and game time, etc. In this research, we explore different fusion strategies (e.g., the fusion on latent features or early-stage features) to form multi-modal data for modeling in order to utilize rich information to improve the result of highlight extraction. Furthermore, we consider the multi-label model based on the factors that affect the highlights to extract the joint features, and combine with the multi-modal model to further improve the final performance. The experiment results show the outstanding performance of the proposed method on the task of sports highlight extraction. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|