All Commentary by AI
Qiyao Sun, Ziqing Li, Ruize Ma, Yuxuan Wang, Zheyun Zhao, Xiongce Lv · 2024
Small-scale sports events have gained widespread popularity through social media platforms, but the lack of professional commentators leads to suboptimal viewer experiences.Traditional automated commentary methods lack deep understanding of game events and generate content that lacks vividness and cultural adaptability.To address this issue, we propose an end-to-end automated basketball commentary system.We combine You Only Look Once(YOLO) object detection and multi-object tracking techniques to perform real-time detection and tracking of basketballs, players, and hoops.By setting fixed regions at hoop positions and detecting whether the basketball's trajectory crosses these regions, we achieve automatic score recognition.Using human pose estimation, we conduct temporal analysis of players' keypoints and accurately identify violations such as holding, traveling, and double dribbling based on rule-based judgments.Compared to methods relying on complex action classification models, our approach reduces dependence on large-scale training data and improves recognition efficiency.Finally, we integrate prompt engineering with large language models to generate real-time commentary that aligns with the audience's language habits and cultural backgrounds.Experimental results show that audience satisfaction increased by 21% compared to a baseline model without style adjustments.This study provides an effective and innovative solution for the automated commentary of small-scale sports events.