基于CNN-BiGRU的足球视频片段分类方法

A Classification Method of Soccer Video Based on CNN-BiGRU

摘要: 基于深度学习的视频分类是体育视频研究的一个重要方向。针对目前视频事件类型识别率低的问题，本文提出了一种基于CNN-BiGRU网络的足球视频事件分类方法。该方法首先利用PySceneDetect工具的场景切换检测功能对完整足球视频进行镜头分割，在此基础上构建包含五类足球事件的数据集；随后通过实验对比，选择将目前主流的卷积神经网络VGG16与BiGRU结合构建分类模型。实验结果表明，CNN与RNN的结合，解决了视频中时间维度利用不足的问题，更有效的整合足球视频中时间维度和空间维度的动态信息，实现比传统技术更高的精度和更快的速度。目前该模型对足球视频数据集上的某单一事件识别率最高达到97.4%。

Abstract: Video classification based on deep learning is an important direction of sports video research. Aiming at the problem of low recognition rate of video event types, this paper proposes a football video event classification method based on CNN-BiGRU network. It first uses the scene switching detection function of PySceneDetect tool to segment the complete football video, and builds a data set containing five types of football events on this basis, then, through experimental comparisons, combine the current mainstream convolutional neural network VGG16 with BiGRU to construct a classification model. The experimental results show that the combination of CNN and RNN solves the problem of insufficient utilization of the time dimension of videos, more effectively integrates the dynamic information of two dimensions of time and space in football videos, and achieves higher accuracy and faster speed than traditional technologies. At present, the model has a maximum recognition rate of 97.4% for a single event on the football video dataset.