摘要
采用多个Kinect从不同角度同时捕获场景,将它们的深度图和彩色图结合在一起,通过数据预处理、顶点构建、点云注册和表面重建等步骤得到场景三维模型.整个流程均在GPU上实现以加速运算,实现了基于GPU的迭代最近点算法、基于GPU的八叉树构建、基于有向距离函数的表面重建等关键算法.试验中,整个算法运行帧率达到8.74f/s;重建分辨率达到约5.9 mm.试验表明,算法基本满足实时动态场景重建的要求,重建模型的精度满足非精确计算类应用的需求.
Multiple Kinect devices are used to capture a scene from different views at the same time.Depth and RGB images of the scene are combined to model the scene.The reconstruction process mainly includes data preprocessing,vertex construction,point clouds registration and surface reconstruction.All processes are GPU implemented to speed up the calculating.Key algorithms implemented in this subject include iterative closest point based on GPU,octree construction based on GPU,surface reconstruction based on the signed distance function.Experimental results show that the algorithm efficiency reaches 8.74f/s and the resolution of models reconstructed are about 5.9 mm.The experimental results prove that,the reconstruction process implemented meets the requirements of the real-time dynamic scene reconstruction and the precision of reconstructed models meets the need of non-exact calculation application.
出处
《东华大学学报（自然科学版）》
CAS
CSCD
北大核心
2015年第4期448-454,共7页
Journal of Donghua University(Natural Science)
基金
国家自然科学基金资助项目(61173105
61373085)
