Near-Infrared Colorization using Neural Networks for In-Cabin Enhanced Video Conferencing

Classical color pictures can not be captured in every scenario, due to the light condition limitations, especially when driving during the night. In those situations, cameras that capture the Near-Infrared spectrum of frequencies prevail. However, pictures from this domain have numerous differences to the ones in the visible spectrum. Therefore, processing Near-Infrared images is a task that requires extensive consideration. In this paper, a method for Near-Infrared image segmentation using artificial neural networks is proposed, together with a colorization algorithm and a novel pipeline for obtaining real-time realistic RGB images, with virtual background for in-cabin enhanced video conferencing.