基于GANs的夜晚场景图像语义理解

Recently, with the increasing demand of growing market of ADAS and autonomous car, the technology of CV is dramatically developing.However, the comprehension of complicated scene image is still a difficult problem. This paper focus on the comprehension of night traffic scene image. Different from the traditional way that processes the night image directly, we propose a new method to comprehend the image regarding the feature of night image. This paper crawled near 30,000 pairs ofday and nightimages from AMOS webcams,then filtered and processed them to meet the needs. This paper trained the GAN model with the input of night image and output of day image. After that, the paper segmented the image, a day image that is generated froma night image, as the segmentation result using SegNet. The features in the method can also be used in scene classification. This paperfinally got the night images\' sematic meanings.