Technique for automated target object search in video stream from UAV in post-processing mode

Abstract

Data obtaining tasks for correct and relevant management decision making pusposes are important. Particularly, in military and rescue areas such data is videodata from an unmanned aircraft vehicle (UAV) camera obtained during its flight over a territory-of-interest. In this case large size of obtained data means a significant problem because of complicating of its manual processing by operator (expert). In addition, data availability must be provided. In practice, the mentioned task is usually solved by recording target videodata onboard during the UAV flight followed by recorded videodata processing after the UAV landing e.g. on the groung control station, i.e. in offline mode. It is obviously to see that using this technique doesn’t solve the problem of complicated target data processing due to manual approach. As for automation of target data processing, as practice shows, every object detection method can potentially decrease processing time, but cannot increase processing quality in comparison with manual processing by operator (expert). Thus, task of ensuring an appropriate balance between availability of target data (videodata from UAV), automation and quality of its processing is relevant. This article i) proposes the technique for automated target object search in videodata from reconnaissance UAVs in post-processing mode by using an adaptive suspicious object search method as an automatic part, ii) describes the corresponding program implementation on C++ for detection method, C# for the user interface part and [standard] platform invoke technique for using the first code (C++) inside the last (C#), iii) shows quantitative characteristics calculated on the set of test videodata. The proposed technique is considered as an appropriate way to solve the specified task.