中文摘要：

提出了一种新的获取出手帧的处理算法。首先,通过在运动员的投篮手臂的关节点处进行颜色标记,以方便提取手臂关节点的特征。然后,通过高斯背景建模进行前景提取,并在此基础上通过标记特征,进一步获取标记点的轮廓,再通过对轮廓进行统计分析获取关节点的坐标信息。最后,对标记关节点的竖直坐标的轨迹进行分析,并通过数学模型的求解以准确获取出手帧。文中所提算法通过使用单个摄像机记录罚球的动作视频,从而克服传统方式中必须以人工回放每一帧图片所产生的局限性。而且,模型参数可以通过自适应选择,使得该算法不依赖于数据源的参数限制。通过大量的真实场景实验,证明了所提算法的有效性和鲁棒性。

英文摘要：

In this paper,a novel algorithm is presented for the prediction of.the free throw.First,three typical joints of the arm are marked.Then,a Gaussian mixture model is used to extract the foreground targets.Some morphological operations are further applied to these targets for more accurate descriptions.After the contour of each marker is generated,the centroid of each joint can be obtained.Finally,the tracks of each marker derived from a series of frames are acquired.The parameters of the proposed method can be automatically estimated,and thus our method can still work even if the parameters of video source are unavailable.