基于模糊熵高斯聚类的弱点状动目标跟踪技术

Algorithm based on maximum fuzzy entropy Gaussian clustering for tracking dim moving point target

研究了一种基于最大模糊熵高斯聚类的实时图像目标跟踪算法:在目标初始信息(位置、速度)已知的情况下,应用最大模糊熵高斯聚类的方法进行跟踪窗内测量点融合,将融合后的点输入到Kalman滤波器中进行预测目标点下一个状态的位置,在预测位置继续开一个跟踪窗进行检测、融合,直至所有图像都被跟踪完为止。理论及实验结果表明,在序列图像情况下该算法能够在保持跟踪实时性的同时,提供较高的跟踪精度。

A new algorithm is described for real-time tracking in image sequences based on the maximum fuzzy entropy Gaussian clustering： In the case that the initial information（position, velocity） of the target are known, it adopts the maximum fuzzy entropy Gaussian clustering to fuse the measured points in tracking window,then the fused point will be inputted to the Kalman filter to predict the next state,then continue creating a tracking window at the predicted state,detect the possible points, fuse them,until all the images have been tracked.Theoretical and experimental results show that in image sequence this algorithm can keep track of real-time,while providing high tracking accuracy.