基于贝叶斯分割和灰预测的人运动跟踪

Human body motion tracking based on Bayesian segmentation and grey prediction

在人运动的视觉分析中,根据差分图像的直方图分布,将目标区域和背景区域作为2个类别进行判别,提出基于最小错误率的贝叶斯决策的动态图像分割方法,获得了良好的分割效果。提出了基于改进的灰预测模型GM(1,1)的人运动跟踪方法,GM(1,1)的初始信息由c均值聚类结果提供,同时GM(1,1)的预测结果作为下一帧图像c均值的初始聚类中心,提高了系统的实时性。与α-β-γ滤波的跟踪误差对比实验证明:该方法能够更好地挖掘人的当前运动规律,能够稳定地保持较小的跟踪误差,从而更好地反映人运动趋势,快速准确地预测人的运动位置。

In the vision analysis of human body motion, taking object region and background region as two different judging categories according to histogram distribution of difference image, this paper puts forward a dynamic image segmentation method based on Bayesian decision with minimal error ratio, which obtains fine segmentation effects. A human body motion tracking method is proposed, which is based on modified grey prediction model GM （ 1,1 ） whose initial information is provided by the results from c-mean clustering. Moreover, the prediction result of model GM （ 1, 1 ） is taken as the initial clustering center for the next frame to improve the real time performance of the system. Experiments on tracking error indicate that compared with α-β-γ filter, this method can explore current law of human body motion better and maintain smaller tracking error stably, and therefore, better reflect the trend of human body motion so as to predict the position of human body motion quickly and correctly.