Fast Fog Detection for De-Fogging of Road Driving Images

Abstract

SUMMARYAdvanced driver assistance system (ADAS) can recognize traffic signals, vehicles, pedestrians, and so on all over the vehicle. However, because the ADAS is based on images taken in an outdoor environment, it is susceptible to ambient weather such as fog. So, preprocessing such as de-fog and de-hazing techniques is required to prevent degradation of object recognition performance due to decreased visibility. But, if such a fog removal technique is applied in an environment where there is little or no fog, the visual quality may be deteriorated due to excessive contrast improvement. And in foggy road environments, typical fog removal algorithms suffer from color distortion. In this paper, we propose a temporal filter-based fog detection algorithm to selectively apply de-fogging method only in the presence of fog. We also propose a method to avoid color distortion by detecting the sky region and applying different methods to the sky region and the non-sky region. Experimental results show that in the actual images, the proposed algorithm shows an average of more than 97% fog detection accuracy, and improves subjective image quality of existing de-fogging algorithms. In addition, the proposed algorithm shows very fast computation time of less than 0.1ms per frame.

Unsupervised Contrastive Masking for Visual Haze Classification

“…Threshold-based methods need to model the haze functions by theory or observation and then substitute statistical or image processing information to determine the class according to thresholds [1]. The information includes the lowest/highest pixel value of the original RGB image [9,18], dark channel priors [7], depth map [14], transmitted image [3], etc., through logarithm, division, and pooling calculation parameters for operations such as transformation [14]. These methods are limited by the construction of specific functions and have poor scalability.…”

“…• Threshold based methods: There are three algorithms using threshold based feature including Filter-Based Fog Detection [9], Saturation & RGB-correlation Detection [1] and HSV-Based Fog Detection [18]. These methods use pixel value information to calculate the haze concentration by the formulas.…”

“…This section presents a performance comparison between MagNet and existing haze classification methods, including three thresholdbased features methods: Saturation & RGB-correlation Detection (S&RCD) [1], Filter-Based Fog Detection (FBFD) [9], HSV-Based Fog Detection (HBFD) [18], the handcrafted features mathod SVM [31], and the Deep Learning features method CNN [27], LeNet5 [10] and ResNet18 [8]. For both algorithms, we fine-tune their hyperparameters to obtain their best performance in the experiments.…”

“…Existing methods for haze classification lie in three main categories, including threshold-based models, handcrafted-featurebased classifiers, and deep learning algorithms. The threshold-based models develop mathematical equations to compute the values of haze degrees directly from haze images, such as color [3,9,18] and dark channel prior [7,14]. This line of research heavily depends on observation data and the computation may be cumbersome, leading to an unstable performance to classifying data from different domains.…”

Improvement of Driver Night Vision in Foggy Environments by Structured Light Projection

“…Numerous publications focus on image processing algorithms to obtain clearer images from foggy photographs in order to detect obstacles [1,26]. We have found algorithms to detect photographs of scenes immersed in fog [27] and thus guarantee a more adequate application of the processing algorithms.…”