Факультет інформатики

Permanent URI for this collection

https://ekmair.ukma.edu.ua/handle/123456789/17448

Browse

Browsing Факультет інформатики by Author "Kurochkin, Andrew"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Design and Development of a 3D-Printed UAV
    (2025) Hrynchuk, Tymofii; Kurochkin, Andrew
    Over the last decade, FPV (First-Person View) drones have become a major part of modern culture. Thanks to affordable technologies for manufacturing drone components, they are now widely used in delivery, rescue missions, military operations, and many other fields. Similarly, 3D printing technology has transformed prototyping and small-scale manufacturing — it is now difficult to imagine modern product development without using a 3D printer, whether the part is made from plastic, metal, or composite materials. In this study, I developed, assembled, and tested the physical characteristics of FPV drones, with most of the frame parts printed on a 3D printer. After a series of tests, we compared flight performance, structural strength, and noise levels across different frame designs. As a result, 3D printed drones, in some cases, can be a reliable alternative to traditional composite-based frames. Although they are sometimes more complex and have a lower impact survival rate, they are cheaper and can be easily customized for various purposes. A demonstration video of the test flights is available at: YouTube
  • Thumbnail Image
    Item
    Real-Time Object Tracking Algorithms for UAV Companion Computers
    (2025) Matsevytyi, Andrii; Kurochkin, Andrew
    Integration of object tracking capabilities into unmanned aerial vehicles (UAVs) represents one of the most relevant and at the same time one of the most significant technological challenges when designing autonomous systems. This report examines best industry real-time object tracking algorithms, with respect to UAV companion computers environment and use-case specifics, evaluates their efficiency and performance, and introduces own relevant metrics for comprehensive purpose-fit evaluation of tracking algorithms. The research focuses on comparison between traditional mathematical approaches and modern deep learning-based methods, particularly Siamese network architectures, with motivation to determine optimal solutions for resource-constrained UAV companion computers. During the analysis, the UAV123 dataset was used, 9 tracking algorithms were overviewed and extensively studied. 5 of them were selected, implemented or deployed, and evaluated: Lucas-Kanade Tracker (KLT), Minimum Output Sum of Squared Error (MOSSE), Discriminative Correlation Filter with Channel and Spatial Reliability (CSRT), Distractor-aware Siamese Region Proposal Network (DaSiamRPN), and NanoTrack. In this studies we found out that deep learning-based trackers significantly outperform traditional approaches in tracking accuracy forUAV related use cases. However, computational requirements vary across platforms, with DaSiamRPN requiring GPU acceleration for real-time performance while NanoTrack maintains reasonable frame rates even on CPU-only platforms. Clear guidelines for tracker selection based on UAV class and mission requirements were established.