基于拉格朗日高斯变换的奥运会视频分类机制研究

随着现代体育不断发展,奥运会承办比赛种类越 来越繁杂,对赛事视频分类提出了一 个新的挑战。现有的人工分类方法无法有效地区分团体竞技类比赛(球类)和个人竞技类比 赛(田径类)视频,从而进行大规模自动分类存储。然而,为了有效地重复使用这些视频文 件,需要对其进行分类存储,主要目的在于提高资源的利用率。针对人工分类手段太过于低 效的现状,本文对奥运会运动视屏内容分类问题进行研究,并提出了一种基于关键帧特征提 取和支持向量机(Supported Vector Machine,SVM)的视频分类方法。以第31届奥运会的 体育视频作为数据集,对每个视频进行关键帧提取和总结,并借由拉格朗日-高斯变换来计 算视频对应的特征向量,将特征向量作为SVM分类器的输入进行体育视频分类。实验结果表 明,对于任意奥运视频,提出的方法平均能够取得70%以上的正确分类率,而错误分类的比 例始终低于10%。特别地,对于奥运中的射击类视频,平均正确分类 率接近90%左右,说明了提出方法的有效性。

Vedio shot boundary detection method based on 3D denseNet

With the continuous development of modern sports,the types of Olympic Games are becoming more and more complex,which poses a new challenge to the cl assification of sports video.The existing manual classification methods cannot effectively distinguish between group sports (ball games) and individual sports (track and field) videos,so as to carry out large-scale automatic classificati o n storage.However,in order to effectively reuse these video files,it is neces sary to classify and store them.The main purpose is to improve the utilization of resources.Aiming at the low efficiency of manual classification methods,thi s paper studies the content classification of Olympic sports video,and proposes a video classification method based on key frame feature extraction and support vector machine (SVM).With the sports video of the 31st Olympic Games as the da ta set,the key frames of each video are extracted and summarized,and the corre sponding feature vectors of the video are calculated by Lagrange-Gauss transfor m .The feature vectors are used as the input of the SVM classifier to classify th e sports video.The experimental results show that for any Olympic video,the pr oposed method can achieve an average correct classification rate of more than 70%,while the error classification rate is always less than 10%.In particular,f or shooting videos in the Olympic Games,the average correct classification rate approaches to 90%,which shows the effectiveness of the proposed method.