Performance Analysis of Fog Removal Models Based on Fog Density Using Synthetic Fog Images

Adverse weather conditions such as fog can severely impair object detection performance in autonomous driving systems. Therefore, research on improving visibility through efficient fog removal is being conducted. However, excessive fog removal can result in losing image details or introducing artificial artifacts, while insufficient fog removal may obscure important objects. This study compares various synthetic fog image techniques and analyzes the performance of fog removal models trained with different fog densities to evaluate their performance based on fog density. Models trained with light fog show significant performance degradation, while those trained with dense fog experience less performance loss. In the future, we plan to develop a fog density estimation algorithm for optimal fog removal and will compare the performance of various fog removal algorithms to maximize their effectiveness.