顾及轨迹地理方向的监控视频浓缩方法

针对视频浓缩研究未引入运动目标轨迹地理方向进行分析，所实现的可视化效果不能区分表达不同轨迹方向的运动目标的问题，提出了顾及轨迹地理方向的监控视频浓缩方法。该方法使用单应模型将运动目标轨迹由图像空间映射至地理空间，利用属性阈值聚类（quality threshold clustering，QT）算法对轨迹出入口进行聚类，实现轨迹地理方向分析。在对轨迹进行映射聚类处理的基础上，通过选取虚拟场景视点实现视频浓缩背景构建，生成轨迹类拟合中心线作为运动目标显示轨迹，并通过构建轨迹类表达模型实现运动目标显示排序，最终实现视频浓缩构建。实验结果表明，该方法能依据轨迹地理方向对运动目标进行区分表达，并能灵活调整不同轨迹类的表达策略。

Surveillance Video Synopsis Considering Trajectory Geographic Direction

In the existing studies of video synopsis, the lack of trajectory geographic direction and partition expression forthe moving objects' trajectories is the key problem yet to be solved.A novel approach for surveillance video synopsis was proposed, in which trajectory geographic direction was considered.The homography model was used for mapping the trajectory from image space to geographic space, and the trajectory analyzation of geographic direction was carried out by clustering the Trajectories. On the basis of mapping and clustering, the video synopsis background was built by selecting the virtual scene viewpoint, and moving objects' display trajectory was determined by creating the fitting centerline, and the expressing order was determined by creating the expressing model of the cluster. Finally the video synopsis was completed. By conducting several experiments, the effectiveness and flexibility of the approach is validated.