基于多信息融合的可驾驶道路检测算法  

Drivable road detection algorithm based on multi-information fusion

作　　者：朱振文 周莉[1] 刘建[1] 陈杰[1] ZHU Zhen-wen ZHOU Li LIU Jian CHEN Jie(R＆D Center for Green Energy Automotive Electronics, Institute of Microelectronics of Chinese Academy of Sciences,Beijing 100029, China School of Electronic, Electrical and Communication Engineering,University of Chinese Academy of Sciences, Beijing 100049, China)

机构地区：[1]中国科学院微电子研究所新能源汽车电子研发中心,北京100029 [2]中国科学院大学电子电气与通信工程学院,北京100049

出　　处：《计算机工程与设计》2017年第5期1389-1393,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(61234003;61304202)

摘　　要：为增强高级辅助驾驶系统中对城市复杂路况的可驾驶道路检测性能,提出一种稳健的可驾驶道路检测算法。对彩色图像,通过特征空间变换去除光照强度变化、阴影等因素的影响,提取图像的光照不变特征,在线建立道路模型并分类得到道路区域二值图,对道路区域二值图进行滤波处理得到道路的相似度分布;将相似度分布和经验性道路空间先验分布相结合进行贝叶斯概率分析,获得可驾驶道路的感兴趣区域;利用感兴趣区域内的深度信息,通过RANSAC算法排除非道路点,获得可驾驶道路区域。在KITTI数据集上的实验测试结果表明了该算法的有效性和鲁棒性。To enhance the performance of drivable road detection in advanced driver assistance systems （ADAS） in complex ur-ban road environment, a robust road detection algorithm was proposed. For color image, its illuminant-invariant feature was ex-tracted after removing the influence of illuminant variance and shadows in the way of the feature-space transformation, which was used for building the model-based classifier. Likelihood distribution was computed by filtering the binary images, which was ob-tained using the model-based classifier. Road region of interest （Road-ROI） was estimated by Bayesian probability analysis com-bining likelihood distribution and empirical spatial priors distribution. Drivable road region was generated using depth information in Road-ROI and RANSAC to filter out the non-road pixels. Experimental results on KITTI-Road datasets verify the effective-ness and robustness of the proposed method

关 键 词：可驾驶道路区域检测 光照不变性 计算机视觉 贝叶斯 深度信息 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]