RGB Guided Thermal Super-Resolution Enhancement

In visual surveillance and security problems, objects can occur in different conditions of illumination and occlusion, therefore thermal images have become a major tool in a large variety of applications. By the reason of their high cost compared to their visual (RGB) counterpart, thermal sensors are used in low-resolution and in low contrast which introduces the necessity to obtain a higher resolution version. In this work, we propose a deep learning model by which to enhance the thermal image resolution guided by RGB images using GAN based model. The results indicate an improvement in resolution enhancement using RGB guided thermal super-resolution models compared to the classical single thermal super-resolution approach.