面向复杂动态场景的多无人机协同摄影

Multi-UAVs Collaborative Cinematography for Complex Dynamic Scenes

当前,对复杂动态场景的实时拍摄仍然以人为操控的方式为主.为了使用无人机对这种场景自动地进行完整拍摄,提出将其分解为场景分割、拍摄构图和无人机调度3个子问题的方法.首先采用带有阈值的K-means聚类算法完整、有效地对场景进行合理分割,以获得一系列小场景,每个小场景对应一架无人机;然后根据小场景之间的位置关系计算获得无人机最佳的拍摄位置,保证在完整拍摄每个小场景的同时,无人机群拍摄的画面有一定的关联性和整体性;最后将无人机飞行调度问题转化为任务指派问题,使用匈牙利算法将拍摄位置指派给每架无人机完成相应的拍摄任务,使无人机群的飞行距离总和最短,完成对复杂动态场景的实时持续拍摄.实验结果表明,该方法可以有效地使用仿真程序控制多架无人机,对复杂动态场景进行多角度且包含全部对象的实时持续自动拍摄.通过观察无人机的飞行轨迹和拍摄结果中目标物体的位置,可以发现,该方法相比场景均分等直接方式,拍摄效果有明显的提升.

For real-time shooting of complex dynamic scenes,people still rely on controlling manually now.For automatically and completely shooting this kind of scene,a feasible solution is proposed,which divides the problem into 3 sub problems:scene segmentation,photography composition,and UAV(unmanned aerial vehicle)scheduling.First,on scene segmentation,K-means clustering is utilized with threshold to get a series of small segmented scenes.Each small scene is corresponded by a single UAV.Then,according to the location of these small scenes,the best shooting positions of UAVs are calculated,which ensure that each small scene is shot completely and meanwhile the pictures shot by different UAVs are relevant and holistic.Finally,in order to minimize the total flight distance and flight cost of the UAV group,the UAV flight scheduling problem is turned into a task assignment problem,where Hungarian algorithm is applied to determine the final shooting positions for each UAV to complete its corresponding shooting task.By this way,a complex dynamic scene can be continuously shot in real-time.Experiments in simulation program shows that the algorithm can provide effective control on multiple UAVs to continuously and automatically shoot a complex dynamic scene,including all the objects inside the scene from multi-view in real-time.By observing trajectories of UAVs and objects’positions in shooting results,it is sure that proposed method produces results with much higher quality comparing with the naive method of average view division.