Deep Learning-Based Human and Vehicle Detection in Drone Videos

Abstract

Nowadays, the detection and tracking of stationary or moving objects have begun to be of great importance for military applications as well as for civilian applications. In this case, it is necessary to use deep learning methodologies in order to effectively meet the emerging needs. This study, it is aimed to detect the people and vehicles in the videos recorded by drones in an environment suitable for field conditions. For this purpose, DarkNet-53 architecture in YOLOv3 was used to detect the presence of people and vehicles in motion in videos with 25 (Frame Per Second) images transferred to the screen in one second. The convolutional neural network has been developed by supporting it with various hyperparameter optimizations and an accuracy rate of 78 percent has been achieved.