In 2020, record heavy snowfall have been occurred owing to climate change. Actually, 2,000 vehicles on the highway could get stuck for three days. Due to the freezing of the road surface, 10 vehicles could have a billiard accident. Road managers are required to provide them immediately in order to alert drivers to snow cover at hazardous location. This paper proposes a deep learning application with CCTV image post-processing to automatically calculate a snow hazard indicator. First, the road surface of hidden region under snow-covered is translated using generative adversarial network, pix2pix. Second, snow-covered and road surface classes are detected using semantic segmentation, DeepLabv3+ under backbone MobileNet. Based on these trained networks, we enable to automatically compute the road to snow rate hazard index how much snow is covered on the road surface. We demonstrate the applied results to 1,000 CCTV snow images on Hokkaido and North Tohoku area in Japan. We mention the usefulness and the practical robustness.