基于视频的雾天驾驶场景及其能见度识别算法研究

Research of Fog Driving Scenarios and Visibility Recognition Algorithm Based on Video

摘要: 获取实时、全面、准确的道路交通场景信息是预防交通事故的重要前提和基本保障，也是实现城市交通智能化的关键。针对雾天驾驶场景及其能见度的识别，传统算法存在复杂度高、鲁棒性差的问题，且多为固定场景下的识别，很难应用于移动的驾驶场景下。本文提出了一种基于单目视觉的雾天识别和能见度估计算法。该算法以柯什米德定律为基础，通过限定Hough变换的极角、半径，压缩投票空间，减少了计算量和复杂度。自定义的区域增长条件，较好的解决了移动场景下的道路分割准确度差的问题。利用加权平均的图像亮度拐点估计方法，能有效地排除干扰，保证拐点估计的准确度。仿真结果表明，算法能实现移动场景下的雾天及能见度识别，精确度、实时性、鲁棒性较好。

Abstract: Obtaining real-time, comprehensive and accurate road traffic information is the important pre-condition and basic guarantee to prevent traffic accidents, and also is the key to realize the urban traffic intelligent. For recognition of fog driving scenarios and visibility, the traditional algorithm has the problems of high complexity, poor robustness, and more using in fixed scene; it is difficult to apply to mobile driving scenarios. This paper proposed a fog and visibility estimation algorithm based on monocular vision. The algorithm, based on the law of Koschmieder, compresses Hough transformation vote space and reduces calculation amount and complexity by limiting polar angle and radius. Custom regional growth solves the problem of poor accuracy in the mobile scenarios’ road segmentation. The weighted average of luminance method which is used in estimation of inflection point can effectively remove interference and ensure accuracy. The simulation results show that the algorithm can realize the recognition of fog and visibility in mobile scenarios with high accuracy, real-time performance and robustness.