Video defogging system based on deep learning

Abstract

The invention designs a real-time video defogging system based on deep learning. An end-to-end image defogging method is adopted, a lightweight convolutional neural network is carried, the transmittance t(x) and the atmospheric light A parameter in the atmospheric scattering model are learned at the same time, and a defogged image is directly generated; the video image is processed by combining aforeground and background segmentation method and a bicubic interpolation algorithm, so that the real-time performance of video processing is further improved; and a foggy image data set is obtained on the NYU2 data set according to a method of generating a foggy image by an atmospheric scattering model, wherein the foggy image data set is used for training and testing. The designed video defogging system is high in transportability, can be embedded into a hardware processor, can be applied to traffic video defogging, unmanned driving and computer vision assisted driving, is remarkable in defogging capacity and good in visual effect, and is an effective video defogging system.

Description

technical field [0001] The invention belongs to the field of computer image processing, and in particular relates to a video defogging method based on deep learning. Background technique [0002] In recent years, technologies such as image processing and computer vision have developed rapidly, and are increasingly used in intelligent transportation systems. In sunny weather, the intelligent transportation system can obtain clear and rich image information, and can detect, track, and identify moving vehicles, which brings convenience to road supervision and greatly improves the driving safety of car owners. However, in severe weather such as fog and haze, the field of vision becomes blurred, the visibility is reduced, and the function of the intelligent transportation system is greatly reduced. It is difficult to obtain accurate image features for inspection and identification of targets; Reduced, has laid a hidden danger for safe driving. It is particularly important to de...

AI Technical Summary

Problems solved by technology

Through the survey, it is found that the research on video defogging is of great significance for the analysis of traffic surveillance video and the intelligent driver assistance system, but the existing research on video defogging is still less and immature, and the residue of defogging is relatively serious. The real-time performance of video defogging is also poor

Images