基于混合高斯模型及粒子滤波的球体跟踪方法

摘要

为提高目标跟踪的准确性和鲁棒性,提出一种结合混合高斯模型和粒子滤波的球类跟踪算法.对视频流提取出图像的每一帧,对每一帧的像素点建立高斯模型,采用初始化模型和更新模型排除噪声干扰,通过阈值的比较提取前景和背景区域;完成目标检测,将检测的前景像素点归为重要性粒子,增加它们的权值.对比均值漂移(Mean-shift)目标跟踪算法,实验结果表明,该算法能使用少量粒子实现球类跟踪,提高目标跟踪的实时性与鲁棒性.%To improve the accuracy and robustness of target tracking,a kind of ball tracking algorithm which combined Gaussian mixture model with particle filter was proposed.The image of each frame was extracted from video streaming,the Gaussian model was set up for pixels of each frame,the model was initialized and updated to eliminate noise interference.The foreground and background regions were extracted through the comparison of the threshold.Detected foreground pixel was regarded as important particle and their weights were increased when target detection was finished.Comparing with Mean-shift target tracking algorithm,experimental results show that the proposed algorithm uses a small amount of particles to realize ball tracking,and improves real-time capability and robustness of target tracking.