Dinamik aralığın derin sinir ağları ile nesne tespitindeki etkisinin analizi

An important problem in computer vision, particularly in object detection, is being able to perceive objects even under challenging illumination conditions. Being robust to such conditions is especially important in applications, such as autonomous driving. Despite the significance of the problem, existing autonomous driving systems use deep object detection networks with low-dynamic range (LDR) images during both the training phase and the testing phase. In this thesis, we investigate whether high-dynamic range (HDR) images can provide better performance for object detection in autonomous driving systems. For this purpose, we provide a comprehensive analysis of the effect of dynamic range on object detection performance. We compare LDR and HDR images on different illumination conditions and show that HDR performs on par with to LDR counterparts when used without pre-processing including normalization and gamma correction. We also show that after applying this certain pre-processing operations, HDR is able achieve on par detection performance with tone-mapped LDR. Moreover, we propose a novel framework to jointly optimize deep-learning-based tone-mapping operators and object detection networks by using a Generative Adversarial approach. Our architecture achieves the best tone-mapping quality score while maintaining a competitive performance to the best classical tone-mapping operator in terms of detection performance.