Fast and accurate car detection in drone-view

Abstract

With the development of drones, aerial images are used in a variety of applications. We propose a way to detect cars in drone-view fast and accurately. For this purpose we proposes a feature called G-ORF for effective feature description. Also we designed a pose classifier and bin-specific weighted Linear Discriminant Analysis (wLDA) classifier for pose classification and binary classification of each pose respectively. Our method showed real-time performance in HD (1280×720) video in PC environment with high detection accuracy.