基于RGB-D数据的目标分割与实时重建方法

OBJECT SEGMENTATION AND REAL-TIME RECONSTRUCTION APPROACH BASED ON RGB-D DATA

应用RGB-D传感器进行三维重建具有速度和成本的优势,并可获取纹理信息,但扫描数据易受干扰。针对有干扰的外部环境,提出一种基于RGB-D数据的目标分割与重建方法。首先为改善输入图像质量,采用多帧叠加的中值滤波对信息缺失的位置进行预处理。接着,利用静态减除和动态3D Mean-shift相结合的方法分割目标,实现对目标物体的实时定位。最后使用一种优化的点到面ICP方法进行配准,并在此基础上利用随机抽样算法对点云的配准进行加速。采用该方法实现的原型系统能够支持含有背景和一定干扰下的小物体配准,且能有效提高三维重建的速度和自动化程度。

3D reconstruction using RGB-D sensor has the advantages in speed and cost,and can capture texture information as well,but the scanned data is prone to interference.This paper presents a RGB-D data-based method for objects segmentation and reconstruction.First, it uses a median filter of multi-frame superimposition to pre-process the positions with information missing to amend the quality of inputted image.Then,it segments the targets by combining the static subtraction and dynamic 3D mean shift to realise the real-time location of targeted objects.Finally,it introduces an optimised point-to-plane ICP algorithm for registration and based on that uses random sampling method to speed up the registration of point cloud.A prototype implemented with this method can support the registrations of small objects with background and certain interference,and can effectively improve 3D reconstruction speed and automation degree as well.