METHODS FOR MINIMIZING THE EFFECT OF FOG ON COMPUTER VISION SYSTEMS

Atmospheric conditions, significantly reduce image quality, making it difficult for computer vision, unmanned transport and video surveillance systems. Existing fog removal methods include classical contrast enhancement algorithms, approaches based on physically based models and neural network methods. However, classical approaches do not account for the nature of distortion, physics-based models lose accuracy in complex scenes, and neural network algorithms can distort information due to dependence on training data. In the testing phase of the methods, the problem is exacerbated by the difficulty of obtaining paired images with and without fog, which makes research in the area of image restoration in unfavourable conditions particularly relevant. This paper gives an overview of existing fog mitigation methods using different principles and evaluates the quality of their performance.