Autonomous Obstacle Detection and Avoidance in Drones

Abstract

Drone technology has recently advanced, and object detection technology is already evolving. These technologies can be used to find illegal immigrants, locate missing people and items, and detect industrial and natural disasters. In this research, we investigate how to improve object detection performance in such cases. Photography was carried out in a setting where it was difficult to identify objects. The experimental data was based on images taken under various situations, such as changing the drone's altitude, shooting pictures in the dark, and so on. In this work, we recommend a way to make YOLO models more effective at detecting objects. To determine the key indicators, we will input the collected data into the CNN model and the YOLO model, respectively. Precision, recall, F-1 score, and mAP are the major metrics of evaluation. Based on the data comparing the CNN model with the YOLO model, an inference will then be drawn.