基于交通图像的能见度检测算法研究

Study of Visibility Detection Algorithm Based on Traffic Image

图像的天空区域灰度易分布不均匀,传统双亮度差法在目标黑体被遮挡时会产生检测误差,提出用区域生长法分割天空区域,并对天空区域像素灰度求均值来改进。传统暗通道先验法也存在天空区域的透射率值偏小的问题,提出用k-means聚类算法自适应分割阈值来准确分割天空区域,并修正其透射率;其中用阈值法估算大气光亮度来提高抗干扰性。能见度检测对比实验验证改进的双亮度差法和改进的暗通道先验法都比其对应的传统算法检测结果更准确、可靠,适用性更强。

When the sky background gray distribution is not uniform and the black object is blocked, the traditional dual differential luminance algorithm causes some error detection. This article proposes to improve this problem, segment the sky region of the image by using region growing algorithm, and average the pixels＇ gray of this region. Traditional dark channel prior also has the problem that the estimated transmission value of the sky region is smaller, than the actual value. An algorithm to adaptive the segmentation threshold using k-means clustering algorithm is proposed to segment the sky region accurately, fix the sky area transmission and use the thresholds to estimation the atmosphere optical brightness value to improve noise immunity. Visibility detection comparative experiment verifies that improved dual differential luminance algorithm and improved dark channel prior are more accurate and reliable, more applicability than corresponding detection results of traditional algorithm.