Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (13): 36-47.DOI: 10.3778/j.issn.1002-8331.2112-0405

• Research Hotspots and Reviews • Previous Articles     Next Articles

Survey of Object Detection Methods Under Adverse Weather Conditions

LIN Meng, ZHOU Gang, YANG Yawei, SHI Jun   

  1. Laboratory of Signal Detection and Processing, School of Information Science and Engineering, Xinjiang University, Urumqi 830046, China
  • Online:2022-07-01 Published:2022-07-01

特殊天气条件下的目标检测方法综述

林猛,周刚,杨亚伟,石军   

  1. 新疆大学 信息科学与工程学院 信号检测与处理重点实验室,乌鲁木齐 830046

Abstract: Driven by deep learning, object detection methods have made great progress in recent years. However, under adverse weather conditions, such as foggy, dust, rain, snow and other weather conditions, the images will be degraded and blurred, which is difficult to extract effective features and cause problems to the later object detection. This paper summarizes the research of domestic and foreign scholars on object detection tasks in adverse weather in recent years, and analyzes it from three aspects:the establishment and use of relevant datasets, the research of image restoration algorithms for adverse weather conditions and their impact on object detection tasks, and the application of domain adaptive methods of transfer learning for object detection. Meanwhile, the paper also compares the experimental results of relevant methods, and puts forward the research focus in the future.

Key words: deep learning, object detection, adverse weather, image restoration, domain adaptation

摘要: 在深度学习的推动下,目标检测方法在近些年取得了很大的进展。但在特殊天气条件下,如在常见的雾霾、沙尘、雨天、雪天等天气条件下拍摄到的图像会退化模糊,难以提取有效特征,从而对后期的目标检测带来巨大的困难。对近年来国内外学者在特殊天气下目标检测的研究进行了总结归纳,从相关数据集的建立和用途、面向特殊天气条件下的图像恢复算法研究及其对目标检测任务的影响、迁移学习中的领域自适应方法在目标检测的应用三个方面进行分析。对相关方法的实验结果进行综合比较,并提出今后的研究重点。

关键词: 深度学习, 目标检测, 特殊天气, 图像恢复, 迁移学习