Analysis and Study of AI-based UAV Control System
Students Name: Senyk Serhii Bohdanovych
Qualification Level: magister
Speciality: Computerized Control Systems and Automatics
Institute: Institute of Computer Technologies, Automation and Metrology
Mode of Study: full
Academic Year: 2023-2024 н.р.
Language of Defence: ukrainian
Abstract: This paper proposes a machine learning model for video image processing and an autonomous UAV flight algorithm based on neural network data. The purpose of this model is to process the video stream from the camera and predict its depth map. The output of this neural network is used in the autonomous flight algorithm of the unmanned aerial vehicle. The essence of the algorithm is to find and avoid obstacles on the flight route. The result of the implemented model and algorithm is a UAV control system based on a convolutional artificial neural network, so such a UAV is capable of autonomous flight, avoiding collision with obstacles. In the first chapter, the terminology of UAVs is considered, and an overview of the use and modern development of UAVs is carried out. The existing methods of training artificial neural networks and the features of using artificial intelligence in current UAVs are analyzed.[1-3] The second section is devoted to analyzing methods and means of UAV autonomous navigation and creating a machine learning model for image processing. For this purpose, the OpenCV, PyTorch libraries and the Python programming language are chosen. A monocular video camera is chosen as the object detection tool. Additionally, the use of convolutional artificial neural networks is justified.[4-6] In the third chapter, a machine learning model is developed for working with the image and creating a depth map. The functional scheme of the UAV control system based on artificial intelligence is presented. The algorithm of its operation is described. The fourth chapter presents a practical implementation of a machine learning model for processing a video image stream and building its depth map. Input preparation, training and validation of this model are performed. The fifth chapter is the economic part, dedicated to the economic feasibility study. The obtained results indicate the economic feasibility of the work. The conclusions reflect the result of the work. They consist of the possibility of using the work to improve the autonomy of UAVs and the need for further research of the system on a real UAV. Keywords: UAV, convolutional neural network, computer vision module, PyTorch, Python. References: 1. Valavanis, K.P.: Advances in Unmanned Aerial Vehicles: State of the Art and the Road to Autonomy, vol. 33. Springer Science & Business Media (2008) 2. Harmat, A., Trentini, M., Sharf, I.: Multi-camera tracking and mapping for unmanned aerial vehicles in unstructured environments. J. Intell. Robot. Syst., 1–27 (2014) 3. Engel, J., Sturm, J., Cremers, D.: Scale-aware navigation of a low-cost quadrocopter with a monocular camera. Robot. Autonom. Syst. 62(11), 1646–1656 (2014) 4. Ravi Garg, Vijay Kumar B. G, and Ian D. Reid, Unsupervised CNN for single view depth estimation: Geometry to the rescue, 2016 – 173 c. 5. Huang, H.-M.: Autonomy levels for unmanned systems (ALFUS) framework, volume I: Terminology, Version 2.0 (2008) 6. Richard Szeliski (2010). Computer Vision: Algorithms and Applications. Springer-Verlag. ISBN 978-1848829343.