Journal
VISUAL COMPUTER
Volume 39, Issue 9, Pages 3905-3922Publisher
SPRINGER
DOI: 10.1007/s00371-022-02534-x
Keywords
Dataset; Deep learning; Fog; Haze; Unpaired dataset
Categories
Ask authors/readers for more resources
Dataset plays a crucial role in designing a defogging/dehazing framework for road safety. Existing synthetic datasets fail to adapt well to real-world scenarios, thus a new dataset containing real-roadside images is proposed in this paper. Experimental results reveal that the quality and performance of dehazed images depend on the dataset used.
Dataset is an important aspect for designing a defogging/dehazing framework for road safety. Most of the datasets available in the literature are synthetic in nature, i.e., fog/haze is added to the clear images. The algorithms trained and tested with these synthetic databases work differently in the real-time scenario due to the presence of uncertainty such as density of fog in the real world. In this paper, a new dataset with real-world hazy and clear unpaired images for road safety called Hazy Unpaired Dataset for Road Safety (HUDRS) is presented. HUDRS consists of thousands of foggy and fog-free real-world roadside scenes captured under the natural environmental conditions. These images are captured from Canon Power ShotSX400 IS camera. The performance of the existing and proposed datasets has been evaluated by using prior-based and deep-learning-based dehazing algorithms present in the literature. Experimental results reveal that the visual quality and performance metrics of dehazed images obtained by implementing a visibility restoration algorithm depend on the dataset on which they are trained or validated.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available