监督学习算法的视频动作识别改进研究

Research on athlete's action recognition in sports video based on supervised learning algorithm

体育视频图像质量差、摄像机非静止、队员图像分辨率低等问题是目前体育视频处理的难点,文章在国内外研究的基础上,基于支持向量机分类算法和运动描述符计算,提出了可用于摄影录像中运动员动作分析的有监督学习算法。根据比赛场地光流场的空间分布特征,利用栅格划分的方法来对视频中运动员的细节动作进行识别,并计算其描述符,最后使用支持向量机的监督学习算法作为模式分类器,识别队员动作的类型。通过对文中算法进行实验验证和分析,表明该算法在很大程度上提升了光流特征的鲁棒性,能够相对非常有效地对广播体育视频中的队员动作进行识别,对以后的动作识别研究具有理论参考价值。

The poor quality of the sports video,the camera still image and the low resolution of the athlete image are the current sports video processing problems. This paper referred to the researches at home and abroad, based on support vector machine classification algorithm and motion descriptor calculation,a supervised learning algorithm is proposed to recognize action of athletes in the sports video.According to the distribution characteristics of the venue space optical flow,using the method of grid division to the video player the details of action recognition,and it calculates the descriptors,finally using supervised learning algorithms of SVM as classifier it recognizes the types of athletes ＇actions.Through the experimental verification and analysis,the high resolution of this proposed algorithm is demonstrated and the results show that this algorithm can improve the robustness of optical flow to a great extent. It is reference value to future research on action recognition.