摘要
室内服务机器人能否快速准确地搜寻到物品是完成物品传递任务的关键.由于待搜寻物品位置的不确定性,造成远距离时难以准确且实时获取物品语义和位置信息.针对此问题,模拟人对环境的分层次感知并结合物品间的位置结构信息,提出一个基于视觉的根据标识物品定位归属(搜寻)物品的实时识别与定位系统.首先采用PCA和质心坐标法提取场景中地面,利用地面凸包和改进的区域生长分割算法实现自下而上的聚类;然后利用CVFH特征在标识库中完成对标识物品的识别,再根据返回的质心坐标近距离获取点云图.重复上述步骤,在依据标识归属关系选取的子归属库中完成归属(搜寻)物品的识别与定位.在"Turtlebot"服务机器人实验平台下的实验结果表明,该系统是有效的.
Whether or not an indoor service robot can quickly and accurately search for objects is the key to completing the objects delivery task.Due to the location uncertainty of the objects to be searched,it is difficult to obtain the objects semantic and location information accurately and in real time at a long distance.To solve the problem,this paper proposes a vision-based real-time identification and location system that locates belonging objects by identification objects by simulating human’s hierarchical perception of environment and combining location structure information between objects.Firstly,extracting ground in the scene by using the PCA and centroid coordinate,using the ground convex hull and improved region growth segmentation algorithm to achieve bottom-up clustering.Then the CVFH features are used to identify the identification objects in the identification database.After using the returned centroid to obtain the point cloud image for near distance,repeating the above steps to identify and locate the belonging objects in the child belonging database selected based on the identification belonging relationship.At last,experiment and analysis are carried out on the“Turtlebot”service robot platform to verify the effectiveness of the algorithm.
作者
李文静
吴皓
田国会
Li Wenjing;Wu Hao;Tian Guohui(School of Control Science and Engineering,Shandong University,Ji’nan 250000)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2018年第12期2335-2343,共9页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61773239)
山东省自然科学基金(ZR201702180022)
山东省科技重大专项(2015ZDXX0101F03)
山东省重大研发计划(2015GGX103034)
关键词
标识归属物品
分割聚类
全局特征
物品识别定位
identification belonging objects
segmentation clustering
global features
object recognition and localization
