摘要
图像序列的运动目标分割检测与跟踪是一个具有广泛应用的困难问题,特别当目标被遮挡丢失和/或大幅度机动时,算法要有很强的鲁棒性才能实现高精度的连续跟踪。通过特征角点提取、由粗到细层级匹配特征光流及其聚类分析算法实现运动目标分割,利用卡尔曼滤波估计对目标的多特征点构成的最小凸多边形描述子进行跟踪,具有抗遮挡丢失和抗机动功能,因此在跟踪过程中能够较好地保持或重新检测目标的最小凸多边形,使检测与跟踪算法鲁棒性强。对运动目标真实图像序列的实验结果充分证明了算法的优良性能。
It is a difficult problem to segmentation, detect and track moving target with an image sequence, which have abroad applications. Especially, when the target is sheltered and/or makes significant motion, the algorithm only with strong robusticity has capacity to complete precision and continual tracking. The segmentation of moving targets is completed by the algorithms of pick-up of feature corner points and detection of feature optical flow using a hierarchical coarse-to-fine matching as well as clustering analysis. The least protruding polygon descriptor constituted by many feature points of the detected target is tracked by using Kalman filtering estimations, which has the functions of resisting lose begotten by shelter and mechanization. Therefore, the least protruding polygon of target can been kept better or detected afresh, which make the robust of detecting and tracking algorithms very strong. The experiment results obtained with real image sequences included moving target have sufficiently proved the choiceness performances of the algorithms proposed.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2005年第3期422-426,共5页
Systems Engineering and Electronics
关键词
运动目标
光流
聚类分析
分割
卡尔曼滤波
跟踪
moving target
optical flow
clustering analysis
segmentation
Kalman filtering
tracking
