Scholarlink

<p>With the extensive popularity of baseball, fans are eager to relive exciting moments such as strikeouts, catching, and home runs. However, manually extracting these highlights from long videos is time-consuming and labor-intensive. To address this, this paper introduces a mechanism called CCM (capturing captivating moments), which aims to effectively identify baseball strikeout highlights. Initially, the proposed CCM employs a coarse-grain policy that involves two key components. Firstly, it utilizes You Only Look Once (YOLO) to detect the change of out-indicator in video frames. Secondly, it employs long short-term memory to analyze the skeleton features of athletes, aiming to extract the draft segments as the candidates that contain the strikeout. Then a fine-grain policy integrates YOLOv5, bidirectional encoder representations from transformers, and 3D convolutional neural networks to accurately identify the strikeout highlights. By combining heterogeneous features and multi-model integration, the proposed CCM ensures robust and precise identification of captivating strikeout moments in baseball videos. The simulation results demonstrate that the proposed CCM outperforms the existing mechanisms in terms of accuracy, recall, precision, and F1-score.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Capturing captivating moments: a multi-model approach for identifying baseball strikeout highlights

  • 期刊论文

  • 发表日期：

    2025年1月23日

查看全文
  • Qiaoyun Zhang,
  • Chih-Yung Chang,
  • Cuijuan Shang,
  • Hsiang-Chuan Chang,
  • Diptendu Sinha Roy

摘要

With the extensive popularity of baseball, fans are eager to relive exciting moments such as strikeouts, catching, and home runs. However, manually extracting these highlights from long videos is time-consuming and labor-intensive. To address this, this paper introduces a mechanism called CCM (capturing captivating moments), which aims to effectively identify baseball strikeout highlights. Initially, the proposed CCM employs a coarse-grain policy that involves two key components. Firstly, it utilizes You Only Look Once (YOLO) to detect the change of out-indicator in video frames. Secondly, it employs long short-term memory to analyze the skeleton features of athletes, aiming to extract the draft segments as the candidates that contain the strikeout. Then a fine-grain policy integrates YOLOv5, bidirectional encoder representations from transformers, and 3D convolutional neural networks to accurately identify the strikeout highlights. By combining heterogeneous features and multi-model integration, the proposed CCM ensures robust and precise identification of captivating strikeout moments in baseball videos. The simulation results demonstrate that the proposed CCM outperforms the existing mechanisms in terms of accuracy, recall, precision, and F1-score.