An Image Enhancement Method for Autonomous Vehicles Driving in Poor Visibility Circumstances

Abstract

In most cases, autonomous vehicles perform well in normal lighting conditions. However, their performance is significantly reduced when faced with poor visibility circumstances because of the lack of detail and information. These situations can easily cause collisions and many dangerous accidents for transportation. Many previous efforts have focused on improving images only in a single case, such as low-light, fog, or dust. This way leads to self-driving automobile systems embedding various image restoration models, increasing computation time, and not meeting real-time processing needs. Therefore, designing an enhancer algorithm restoring input image quality is more imperative than ever before. This paper proposes a method to enhance the image under poor visibility circumstances such as foggy and low light. On top of that, we introduce the defogging algorithm based on contrast energy, entropy, and sharpness characteristics. On the other hand, the inverse of an image captured in a dark environment will be equivalent to a daytime image obtained in a fog atmosphere. Inspired by this interpretation, we adopt the proposed defogging model to enhance images in low-light conditions. Experimental results demonstrated that our method was feasible to execute pre-processing input images for autonomous vehicles driving in poor visibility circumstances.