体素连通性的深度图像拉伸特征提取  

Extrusion feature extraction of range image by voxel directed connectivity

作　　者：赵翠莲[1] 范志坚[1] 孙亮[1] 陈昱[1] 

机构地区：[1]上海大学CIMS和机器人中心,上海200072

出　　处：《光学精密工程》2009年第4期930-936,共7页Optics and Precision Engineering

基　　金：上海市重点学科建设基金资助项目(No.Y0102);上海市教育基金资助项目(No.06Az023)

摘　　要：针对深度图像中的拉伸特征,提出一种基于体素连通性和区域生长的快速提取方法。利用线性八叉树结构建立深度图像数据的体素模型,通过加入八叉树编码向栅格编码快速转换算法,改进了体素邻域的搜索效率,使计算时间可减少50%以上。在此基础上,根据种子体素特定方向上的连通性进行区域生长,然后利用阈值对生长结果进行判断并提取出拉伸特征数据。将该方法应用于实际船舶液舱深度数据提取,并对提取结果和阈值选择进行了讨论。实验结果表明,设定对比精度为10-4mm,提取准确率达到了90%左右,可用于点云切片法进行拉伸重构。An extraction method based on voxel connectivity and region growth is proposed for the extrusion features of range images. A voxel model of range image data is established with linear octree codes, and a fast transforming algorithm from octree codes to grid codes is introduced to improve the efficiency of searching voxel neighbors, thereby the computing time has been reduced by 50%. Then, the region growth is carried out based on the connectivity of the seed voxel in a specific direction, and a threshold is used to judge the growth result to extract the extrusion feature data. Finally, the method is applied to the liquid cabin range image data of an actual ship and the results of the extraction and threshold option show that the extracting accuracy reaches about 90% when setting the comparison precision as 10^-4 mm. The experimental results support a eonclution that the method can be used in a point-cloud cutting method to reconstruct the extrusion features.

关 键 词：深度图像 体素 拉伸特征 连通性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]