基于梯度图像和模板匹配的单目视觉障碍物检测方法  

A monocular vision method based on gradient image and template matching to detect obstacles

作　　者：李俊霖 吴晓波 LI Junlin;WU Xiaobo(Department of Logistics Information and Military Logistics Engineering, PLA Army Service College, Chongqing 401311, China)

机构地区：[1]解放军陆军勤务学院后勤信息与军事物流工程系,重庆401311

出　　处：《自动化与仪器仪表》2019年第4期106-109,共4页Automation & Instrumentation

摘　　要：障碍物检测是自动驾驶、机器人自主导航的核心问题之一。为了检测障碍物,提出一种基于聚类和目标匹配的单目视觉方法。在机器人使用场景下,道路的纹理、色彩具有相似或者一致性,在这样的假设下,对可能的道路点和非道路点进行多次取样,通过聚类的方法找出道路点的特征,然后进行匹配,即可标记出全部的道路点,未被标记的点即为障碍物,从而将障碍物检测出来。匹配时考虑图像梯度,将平均匹配准确度提高到90%以上。Obstacle detection is one of the core issues of autonomous driving and robot autonomous navigation. A monocular vision method is proposes based on clustering and template matching to detect obstacles. In the robot applying scenario, the texture and color of the road are similar or consistent. Under this assumption, the possible road points and non-road points are sampled multiple times, and the characteristics of the road points are found through the clustering method. Matching will mark all the road points, and unmarked points will be obstacles obviously. Considering the image gradient when matching, the average matching accuracy has been improved to over 90%.

关 键 词：聚类 模板匹配 单目视觉 图像梯度 障碍物检测 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]