Investigation of computer vision capabilities in autonomous unmanned aerial vehicles control systems

The development of modern hardware and software has led to a rapid increase in the use of unmanned aerial vehicles, primarily aircraft. One of the promising areas for improving the efficiency of such platforms is the development of autonomous control systems, eliminating the need for human operators. The article presents the results of a study on the possibility of constructing elements of an automatic motion control system based on computer vision for unmanned aerial vehicles. The authors used comparative analysis to evaluate the advantages of two popular instruments: YOLO and SSD. They have also collected data for training the chosen model and tested it in various conditions, described the methodology for creating a training set and presented the model testing results on video images captured by a UAV's camera. The test results confirm that the YOLOv8n model is suitable for object detection on the UAV board using a Raspberry Pi 4 Model B as the single-board hardware platform. The object detection accuracy was from 80% to 90%, with the power consumption of 15-25 watts.