摘要
提出了真实场景中的运动目标轨迹有效性判断与自动聚类方法。利用轨迹长度、坐标值方差及目标相邻两帧运动方向等信息,对轨迹进行了预处理,得到有效的轨迹,然后以其作为样本,计算轨迹之间的空间相似距离,采用K均值聚类法,按轨迹的几何形状完成了轨迹聚类。提出了利用目标运动的起始点及整个运动过程中目标的运动方向信息,采用K均值聚类方法,进一步按目标的运动方向完成了轨迹聚类。两种真实场景的目标轨迹聚类结果证明了该方法的有效性。其研究结果为学习轨迹模式、目标运动轨迹识别、分类、异常检测奠定了基础。
A novel method that can accurately validate and cluster trajectories of the moving objects in real scenes was presented. Firstly, through calculating the length, variance of the coordinates and orientation code of the trajectories, valid trajectories were retained. And then the valid ones were taken as the samples and automatically clustered based on K-means approach using the distance between two trajectories. Moreover, a new method to further cluster the trajectories using the start points of them and the information of orientation during the whole process was presented. The trajectories are effectively clustered in two real scenes. Results can provide efficient evidence for the latter work such as trajectory recognition, classification, and anomaly detection.
出处
《计算机应用研究》
CSCD
北大核心
2007年第4期158-160,169,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(60372085)
陕西省科学技术研究发展计划项目(2003K06-G15)
关键词
轨迹聚类
K均值
轨迹识别
分类
异常检测
trajectory clustering
K-means（KM）
trajectory recognition
classification
anomaly detection
