基于系统辨识的体育运动视频识别

System Identification Based Sports Video Recognition

全自动体育运动识别,通过提取视频光流特性和梯度特征,达到聚类的目的。从每一帧的特征中,通过线性系统辨识工具,学习每个视频片段的运动形状。尽管非钢体的运动序列难于进行学习、比较,但是动态描述方程具有很强的比对能力和鉴别能力。使用它们来识别单个运动或者行为得到较好的效果。

Our goal is to extract the motion signatures of moving human figures to recognize sports activities. We begin with a short sequence of images of a moving figure and then analyze the dynamic structure of these sequences using a system identification approach. The learned dynamics and transition matrices show significant statistical variation across various activities. Therefore, a distance is derived from subspace learning as measurements to recognize individual events and activities by the angels of their motion subspace in spatial-temporal domain.