论文部分内容阅读
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.
Trajectory analysis is the foundation of video surveillance scene understanding, but due to occlusion and other reasons, the tracking process will appear incomplete noise trajectories, leading to inaccurate analysis results.Aiming at these problems, using improved trajectory similarity measure and clustering method to make scene area Segmentation.Firstly, the trajectory is coded, and the distance between trajectories is calculated by using the spatial and velocity features of the trajectory to improve the similarity measure.Secondly, an improved hierarchical clustering algorithm is adopted to take the longest trajectory of this kind as the moving object behavior model Represents the trajectory of spatially similar and similar speed features are divided into the same scene area to get the clustering results in line with the actual situation.The algorithm does not need complex preprocessing or filtering of the trajectory, More reasonable.Finally, under the real scene, the validity and universality of this clustering algorithm are verified.