本文介绍一种复杂背景中目标图像的提取与高精度目标形心跟踪算法。通过对贝叶斯（Ｂａｙｅｓ）风险函数最优化，将图像分成多灰度层次的目标与背景，进而得到多灰度阀值，并考虑目标像素之间的连通性，将图像进行二值化，有效抑制了孤立噪声点对分割的影响。根据分离后二值图像，得到了跟踪窗中目标形心估值误差表达式，进而得到一种将目标跟踪窗口不断向目标形心移动的高精度形心跟踪算法。 This article describes a complex background of the target image extraction and high precision target centroid tracking algorithm. By optimizing the Bayesian risk function, the image is divided into multi-grayscale targets and backgrounds, then multi-grayscale thresholds are obtained, and the images are binarized considering the connectivity between the target pixels. Effectively suppress the impact of isolated noise points on segmentation. According to the binary image after separation, the error expression of the target centroid in the tracking window is obtained, and then a high accuracy centroid tracking algorithm which can move the target tracking window to the centroid of the target is obtained.