一种基于深度信息的障碍物检测方法  

A Method of Obstacle Detection Based on Depth Information

作　　者：杨磊[1] 蔡纪源 任衍允 李俄德 

机构地区：[1]上海大学机电工程与自动化学院上海市电站自动化技术重点实验室,上海200072 [2]大唐陕西府谷能源化工有限责任公司,陕西西安719405

出　　处：《计算机技术与发展》2015年第8期43-47,共5页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61005015);国家第三批博士后特别基金(201003280);上海市青年教师培育计划;上海大学青年教师资助计划资助项目

摘　　要：为增强室内移动机器人障碍物检测和道路提取能力,文中提出了一种基于深度信息的障碍检测方法。首先对深度数据进行滤波处理,填补缺失的数据;然后将深度图转换为视差图,对视差图进行水平和竖直方向投影直方图统计获得U-V视差图;由V视差图得到初步道路信息,进一步用最小二乘法拟合出完整道路平面。对U-V视差图进行两次最大类间方差法(Otsu法)分割,提取出障碍物主要信息,并根据视差关系得到障碍物在世界坐标系中的位置。实验结果表明,使用Kinect可以有效地对地面障碍物进行检测并提取出道路信息,可为室内移动机器人提供良好的导航信息。To enhance obstacle detection and road extraction ability of the indoor mobile robot, present an obstacle detection method based on depth information. Firstly ,process the depth data by filtering and fill up the missing data. Then transform depth map to disparity map and calculate U-V disparity map by horizontal and vertical direction projection histogram statistics. Based on preliminary road information got by V disparity map,can fit a complete road plane by using the least square method. With twice segmentation of the U-V disparity map by Otsu＇ s method,extract the main information of obstacles and obtain the location of obstacle in the world coordinate according to disparity relationship. Experiments show that Kinect can effectively improve the ability of obstacle detection and road information extraction for indoor mobile robot,and provide good navigation information.

关 键 词：深度数据 视差图 最小二乘 最大类间方差法 障碍物检测 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]