A joint cross-modal super-resolution approach for vehicle detection in aerial images (Conference Presentation)

Vehicle detection in aerial imagery is still an open challenge in-spite of some existing deep learning methods. It has become an extremely difficult task due to relatively small target size, variable scale and orientation of targets. Though super-resolution; a technique which learns a mapping between low-resolution (LR) images and their high-resolution (HR) version can resolve this problem, struggle remains while detection takes place at night or in a dark environment. Infrared (IR) imaging become necessary for those visual system. In addition, training data may not be available for many tasks. Hence, we focus on learning a joint approach to generate a translated super-resolved image from a low-resolution source domain to a high-resolution target domain. Our goal is to design a framework for vehicle detection in synthesized super-resolved images. Our proposed network builds on the Generative Adversarial Network …