Разработка алгоритма управления длиннофокусной камерой для распознавания людей в публичных пространствах

Выполнил: Тимченко Даниил Геннадьевич

In recent years with the rise of intelligent systems in a lot of modalities (Computer Vision, Time Series Forecasting, Graph Prediction) new approaches to existing tasks emerge in various fields, however those tasks were only possible to complete by human in the loop. One of such fields is camera tracking, as it turns out, efficient camera tracking is a well-studied task for a task of traversing stationary objects, but it becomes exponentially more difficult for a case with moving targets. We (Timur Khaibrakhmanov and Timchenko Daniil) propose a method for dynamic efficient traversing through camera targets, namely "KNN-Greedy" Algorithm, which is able to greedily choose the next most viable target in the process. This paper includes a lot of technical details for camera modulation and algorithmic approach. Second paper in this series (Completed solely by K.T. Radikovich) also covers simulation of targeted agents movement (we chose to conduct experiments on modeled soccer field environment, to make sure, that movement is non-trivial and requires optimal intelligent controlling).  Overall, the paper provides a baseline for further advancement in the theory of Dynamic Travelling Salesman Problem (DTSP) and researches thoroughly on this subject, as well as providing implementation details with full source code.

__________________________________________________________________________________________________________

Итоговый отчет (PDF, 931 Кб)

Презентация к защите (PDF, 101 Кб)