Automatic Generation of RoboCup Soccer Highlight Videos Using a Large Multimodal Model

Abstract

Translate Article

RoboCup consists of multiple leagues, each with distinct rules and tasks, which makes it difficult for general audiences to understand the matches. To address this issue, we propose a method for automatically generating narrated highlight videos that convey match information in an accessible manner. Specifically, we utilize a large multimodal model to extract and summarize match content from videos, generate descriptive texts, and synthesize narration for inclusion in the final video. Subjective evaluations revealed issues such as mismatches between narration and visuals and insufficient extraction of key scenes. However, the proposed method demonstrated appeal in terms of entertainment value.