Car Localization in Aerial Images Taken from Quadcopter

Abstract

Abstract-Unmanned aerial vehicles (UAV) are popular research platforms that find increasing amount of applications in many areas, such as military, civil, commercial, and entertainment due to their high maneuverability, vertical take-off and landing abilities, and suitability for use in indoor and outdoor spaces. Today, small, and single board computers with very high CPU/processor capacities are developed, and by means of these processors, which will be inserted into unmanned aerial vehicle platforms, many realtime machine vision applications became possible. This study discusses the problem of car localization in aerial images taken from unmanned aerial vehicles. Within this context, a new dataset was created by using quadcopter-type unmanned aerial vehicles and various cameras. Both Polyhedral Conic Classifier and You Only Look Once (YOLO) algorithm, which is currently one of the fastest methods in literature, and uses deep learning architecture, were used to locate the cars in collected images, and the results were compared.