Inadequate road infrastructures in developing countries, such as road lighting, is a real life problem affecting their transportation system. A poor road lighting hinders the ability of drivers to perceive the road conditions as well as its objects which may lead to the accident. We address a problem of synthesizing day-time appearance from the night-time road scene using loss-modified generative adversarial network. Our approach basically makes a balance between domain regularization and human perceptual restoration. By taking the night-time image into our enhanced adversarial network, which employs structural similarity loss, we increase its visibility so that the driver can clearly see the road scene like they are driving during the day-time. Experimental results show the benefit of our approach for enhancing the visibility of the night-time view of the road, measured by the image quality metrics.
Night-to-Day Road Scene Translation Using Generative Adversarial Network with Structural Similarity Loss for Night Driving Safety
Studies Comp.Intelligence
2021-04-11
15 pages
Article/Chapter (Book)
Electronic Resource
English
Elsevier | 1981
Engineering Index Backfile | 1928