面向雾天的双先验兴趣区域车道线识别算法研究  

Study on fog-oriented lane detection method based on dual-heuristic region of interest

作　　者：周劲草[1] 魏朗[1] 乔洁[1] 

机构地区：[1]长安大学汽车学院,陕西西安710064

出　　处：《中国安全科学学报》2016年第8期128-133,共6页China Safety Science Journal

基　　金：国家自然科学基金资助(51278062)

摘　　要：为提高浓雾工况下车道线识别的准确率,开发一种基于改进暗通道优先和双先验兴趣区域(ROI)的雾天车道线识别跟踪算法。首先,依据改进暗通道优先算法对雾天图像静态ROI-I进行去雾处理;再按照Scharr滤波器和大津法得到该区域二值化图像,并通过Hough变换精确提取车道线;以此为基准,用Kalman预测器预测下一帧图像中车道线可能出现的动态ROI-II区域;最后采用改进的去雾算法对ROI-II区域去雾,利用B样条插值对该区域车道线进行拟合。结果表明,该算法在薄雾工况下的识别准确率可达99%,在浓雾工况下也能达到96%。In order to improve the accuracy of lane detection in fog, a new algorithm based on dual-ROI and improved dark channel prior algorithm was proposed. Firstly, the static ROI-I of lane images in fog weather was defogged by improved dark channel prior, then the binary images of road edges were obtained by Scharr operator and Ostu algorithm, road lane was obtained by Hough transformation then. The probable dynamic ROI-II of next frame of lane images was predicted by Kalman operator and was defogged by improved dark channel prior, finally the lane in dynamic ROI-II was extracted by B spline interpolation. Experimental results show that the detection accuracy of this new algorithm was 99% in thin fog and 96% in thick fog.

关 键 词：车道线识别 暗通道优先 图像去雾 兴趣区域(ROI) B样条插值 

分 类 号：X924.4[环境科学与工程—安全科学]