An Optimal DCP-Based Defogging Method of Improving Objects Detection in Adverse Weather

As one of the most important components of practical autonomous driving technology, designing efficient and reliable object detection methods has been the focus of attention in academic and industrial circles for recent years. Fortunately, by taking advantage of the development of deep learning technology, various traffic-related object detection methods based on deep networks have been designed using different feature extraction methods, such as convolutional neural networks (CNNs). However, existing approaches often focus on target detection in fine weather conditions and ignore the negative impact of bad weather on target detection performance. Therefore, in this article, we analyze the problem of target recognition in rainy and foggy weather, and do survey of existing defogging methods. Accordingly, we propose our Dark Channel Prior (DCP)-based image-defogging method to improve the detection efficiency of low-visibility targets in rainy and foggy weather. In addition, we discuss the two most commonly used data sets and how they are generated. Then, by considering the experimental results and the practical application scenarios of autonomous driving techniques, we further discuss the open challenges of designing a practical defogging method for supporting autonomous deriving.