Military networks, and especially tactical networks, have become crucial in conducting military actions. With images and videos being rich sources of information, a primary challenge in military networks is to be able to issue a query over a (large) distributed set of devices to find concrete actions or objects of interest. The problem inherent to this scenario is how to gather this information in an optimal way, depending on the objective, for further processing. To this end, in this work we present three different approaches, with different objectives, for using video analytics to label and collect the right information that is stored on mobile devices. We first present a solution for crowdsourcing devices to label videos that contain objects and actions of interest in a minimum amount of time. We then present a solution for gathering all images that contain objects of interest at a central server where a complete search of images stored on the mobile devices is completed as fast as possible. Finally, we consider the case in which we desire to collect images that contain objects of interest as fast as possible, perhaps at the expense of the overall search time taking longer. In all three cases, we divide processing responsibilities between mobile devices where the videos/images are stored and the edge cloud.

中文翻译：

---

通过军事网络视频分析收集信息的信息质量


军事网络，特别是战术网络，在实施军事行动中变得至关重要。由于图像和视频是丰富的信息源，军事网络的主要挑战是能够对（大型）分布式设备集发出查询，以找到感兴趣的具体动作或对象。这种情况固有的问题是如何根据目标以最佳方式收集这些信息以进行进一步处理。为此，在这项工作中，我们提出了三种具有不同目标的不同方法，用于使用视频分析来标记和收集移动设备上存储的正确信息。我们首先提出了一种众包设备的解决方案，可以在最短的时间内标记包含感兴趣的对象和动作的视频。然后，我们提出了一种解决方案，用于在中央服务器上收集包含感兴趣对象的所有图像，其中尽可能快地完成对移动设备上存储的图像的完整搜索。最后，我们考虑这样一种情况：我们希望尽快收集包含感兴趣对象的图像，这可能会导致整体搜索时间更长。在这三种情况下，我们在存储视频/图像的移动设备和边缘云之间划分处理职责。