Detecção e Classificação de Objetos Presentes em Imagens Aéreas de Drones de Ambientes Urbanos
Guilherme R. Sganderla, Cláudio Roberto Marquetto Maurício, Valéria Nunes dos Santos, Fabiana Frata Furlan Peres · 2021
Through large data sets, it is possible to train and instruct a machine with skills to perform tasks previously performed only by humans. This possibility has become increasingly real with the use of Deep Learning and powerful algorithms that have been developed over time. Among them is YOLO, a Convolutional Neural Network algorithm that allows several uses, including the detection and classification of objects contained in images of urban environments, such as people and vehicles, allowing the identification and location of objects within the images. This work presents a model for detecting and classifying common object classes in urban environments People, Small Vehicles, Medium-Vehicles and Large-Vehicles). For this project we used a combination of 3 datasets of aerial drone images of urban environments (Stanford Drone Dataset, Vision Meets Drone, The Unmanned Aerial Vehicle Benchmark Object Detection and Tracking). The result obtained from the initial training of this YOLO algorithm was an average accuracy of 67.2%.