应用于自动驾驶车辆环境感知系统的去雾技术研究

项目名称： 应用于自动驾驶车辆环境感知系统的去雾技术研究

项目编号： No.61502537

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 郭璠

作者单位： 中南大学

项目金额： 21万元

中文摘要： 现有的图像去雾复原算法大多利用大气散射模型采用整幅图像统一增强的方式，若直接将这些算法运用于自主车环境感知系统中道路图像的处理，除不能满足车辆行驶的实时性外，很可能还会产生近处路面区域（能见度范围内）过增强，而远处区域（能见度范围外）增强不够的问题，导致道路标志线、标志牌、交通灯等交通场景目标图像特征缺失，检测率低，影响行车安全。本项目将根据道路图像的特点，基于能见度概念，研究雾天道路环境下的图像分区实时去雾技术，进一步提高去雾效果，保证自动驾驶的安全性。研究内容主要包括：1）构建基于能见度的道路图像去雾新模型，为雾天道路图像的不同区域分配不同的增强系数以分别进行去雾处理；2）设计基于新模型的去雾算法；3）研究基于能见度检测和交通场景目标检测的道路图像去雾效果客观评价方法；4）设计与实现道路图像去雾软件平台，并运用于自动驾驶车辆环境感知系统以检验其鲁棒性与实时性。

中文关键词： 图像去雾；道路场景；能见度距离；增强系数；去雾效果评价

英文摘要： Most recent existing image defogging algorithms perform the same restoration processes for the whole image regions by using the atmospheric scattering model. If we directly adopt these algorithms to process the road images that captured by the environment perception system of autonomous vehicle, apart from the high time complexity, it may also cause the enhancement to be partial in the near road surface (within the visibility distance), while the far-away area (outside the visibility distance) are less enhanced. In this case, the image features of road marks, road signs, traffic lights and other traffic scene objects may be lost, which may lead to a low detection rate and influence the driving safety of autonomous vehicle. Therefore, this project will study the partitioned real-time defogging technique for foggy road images based on the road image features and the concept of visibility. The technique can further improve the image defogging effect and ensure the safety of automatic driving. The project will mainly include the following aspects. Firstly, a new road image defogging model is constructed based on the visibility distance. The new model can perform different restoration processes by assigning different enhancement coefficients for different regions of foggy road images. Secondly, an image defogging algorithm based on the new model is designed. Thirdly, the defogging effect assessment methods based on the visibility distance detection or the traffic scene object detection are studied. Finally, a software platform of road image defogging is designed and applied to the environment perception system of autonomous vehicle to test the robustness and real time capability of the platform.

英文关键词： image defogging;road scene;visibility distance;enhancement coefficient;defogging effect assessment