.png){kind=icon}
.png){kind=icon}
Return
DHMDL: Dynamically Hashed Multimodal Deep Learning Framework for Racket Video Summarization Using Audio and Visual Markers
.png){kind=icon}
OA
AI
DOI：10.1080/08839514.2025.2462382.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Sports videos are being streamed over a large range of social media platforms, and they always have a huge audience base and viewer history. In order to provide more excitement for the users in watching a completed game, automatic video summarization is an inevitable solution. While sports like soccer, cricket have been the main focus of the video summarization research, little attention has been centered over racket sports. Our proposed dynamically hashed multimodal deep learning (DHMDL) sports video summarization framework fuses excitement scores by utilizing deep learning architectures to extract cues from multi modalities namely commentator voice, spectators' cheers and player's expression and then leverages to generate video segment as highlight by using hash codes mapped to weighted sum of excitement score. Also, the proposed synchronized parallel processing ranking based hash map framed using the merge sorting technique for categorizing the excitement scores is applied in video summarization. The framework is tested on U.S. Open and Wimbledon match videos and the results show superior results against state-of-art techniques with normalized discounted cumulative gain (nDCG) score improved by 2%, positive matching highlight segment identification increased by 20% on YouTube Videos.
Keywords:
FACE RECOGNITION
EVENT DETECTION
SPORTS VIDEO
HIGHLIGHTS
NETWORKS
Journal
R
IF:
13.2
Papers:
155
Citations:
3.2K