Generating Object Proposals for Vehicle Detection in Aerial Images

Vehicle detection in aerial images is an important task in many applications such as screening of large areas or traffic monitoring. In general, classifiers or a cascade of classifiers within a sliding window approach are used to perform vehicle detection. However, sliding window approaches are limited for vehicle detection in a real-time system due to the huge number of windows to classify. To overcome this challenge, several objects proposals methods have been proposed for generating candidate windows in detection frameworks. Impressive results have been achieved on common detection benchmark datasets like Pascal VOC 2007 for a significantly reduced number of candidate windows. However, these datasets, which are used to develop the object proposals methods, exhibit considerably differing characteristics compared to aerial images. In this report, we examine the applicability of such object proposals methods for vehicle detection in aerial images. Therefore, we evaluate the performance of seven state-ofthe-art object proposals methods on the publicly available DLR 3K Munich Vehicle Aerial Image Dataset. Relevant adaptions are highlighted by using the Selective Search method. Finally, the adapted methods are compared to baseline approaches like sliding window