智能监控场景中运动目标轨迹聚类算法

轨迹分析是视频监控场景理解的基础,但由于遮挡等原因,跟踪过程会出现不完整的噪声轨迹,导致分析结果不准确.针对此类问题利用改进的轨迹相似度度量和聚类方法进行场景区域分割.首先,对轨迹进行编码,提出利用轨迹的空间特征和速度方向特征改进相似性度量方法计算轨迹间距离;其次,采用改进的层次聚类算法,以该类最长轨迹作为运动物体行为模式代表,将在空间上接近且具有相似速度特征的轨迹划分为同一场景区域,得到符合实际情况的聚类结果.本算法无需对轨迹进行复杂的预处理或过滤,并且加入速度方向特征使区域划分更加合理.最后,在真实场景下,验证了该聚类算法的有效性和普遍适用性.

Moving object trajectory clustering method in intelligent surveillance video

Trajectory analysis is the basis of scene understanding, however noise trajectories causing by barriers in true surveillance scene will influence the result veracity. A trajectory similarity measure and clustering method to segment a scene into semantic regions were proposed to eliminate the effect causing by noise. First, the trajectory was encoded, and then both the object position and its instantaneous velocity were computed by improved similarity measure method to represent the distance between two trajectories. Then, the improved hierarchical clustering algorithm which chooses the longest trajectory as each cluster representation was applied to cluster trajectories according to different spatial and velocity distributions. In each cluster, trajectories were spatially close, had similar velocities of motion, and represented one type of activity pattern. This algorithm does-t need complex pre-process or filter, and because of adding velocity direction, the scene division is more reasonable. Finally, through experiment in true scene, the results show that the method can distinguish different clusters reasonably and improve the effectiveness of clustering.