A Visibility Estimation Method for Foggy Images Based on Single Image Depth Estimation

Abstract

The invention discloses a foggy image visibility estimation method based on single image depth estimation. The invention directly and accurately estimates the depth information from the original foggy image through the convolutional neural network, only needs image acquisition equipment, and does not need multi-sensor fusion, and the system is simpler. Through a large amount of real foggy image data and fog-free image data, cycleGAN is used to train the foggy generator. Compared with the artificially designed foggy degradation model, cycleGAN can generate image data that is closer to the real foggy day. The lack of network data for estimating depth information in foggy images.

Description

technical field [0001] The invention belongs to the field of image processing, in particular to a foggy image visibility estimation method based on depth estimation of a single image. Background technique [0002] Visibility is the visible distance of the target, which refers to the maximum distance that the outline of the target can be distinguished from the background when the target is observed. Visibility is an important element of meteorological observation. Manual observation visibility, generally refers to the effective horizontal visibility, refers to the maximum horizontal distance of the target that can be seen in more than half of the field of view around the station. Daytime visibility refers to the maximum horizontal distance of a target (black, moderately sized) that a person with normal vision can see and recognize from the sky background under the prevailing weather conditions; nighttime visibility refers to: assuming that the overall lighting is increased t...

AI Technical Summary

Problems solved by technology

[0008] (1) The algorithm based on color features is too simple, there are many prior conditions, the threshold is difficult to grasp, and the stability is not enough

[0009] (2) The method based on multi-sensor is too complex and expensive

[0010] (3) The method based on the dark primary color channel has a theoretical basis, and it is difficult to obtain data directly through the method of directly estimating the visibility through the transmittance; the calculation of the visibility by manually marking the observation point is complicated, requires too much manual intervention, and cannot meet the needs of complex scenes; through the camera The parameters roughly estimate the depth information of the scene with poor accuracy

Images