基于彩色图像引导的RGB-D相机追踪与三维重建

RGB-D camera tracking and 3D reconstruction via color image guiding

为了解决基于RGB-D相机三维重建中相机位姿估算不准确问题,尤其是当扫描时RGB-D相机之间移动距离过大或者存在严重遮挡变化,提出一个新颖和鲁棒的RGB-D相机位姿追踪算法来提高相机追踪的精度并进行高保真度的三维重建。首先,采用一个线性处理(line-process)算法来建立连续RGB-D帧之间更加可信的密集对应,然后利用这些密集对应来对传统的ICP(iterative closest point)方法进行改进,使得RGB-D相机的位姿追踪更加精确。尽管手持RGB-D相机获得的深度图包含明显的噪声,并且相邻帧之间很容易出现抖动,所提出的构建密集对应的算法仍然可以对RGB-D相机进行准确地追踪。另外,该算法还可以在闭环检测中得到有效的应用,并可以有效地减轻相机追踪产生的相机漂移。最重要的是该方法还可以作为一个类似的插件应用到其他的基于ICP的算法中来提高相机追踪的精度。实验结果表明,不管是在公共数据集上还是实时扫描的真实场景上,该方法都有效和鲁棒。

In order to solve the problem of inaccurate estimation of camera pose in RGB-D 3D reconstruction,especially when the moving distance between RGB-D cameras is too large or there are serious occlusion changes during scanning,a novel and robust approach for RGB-D camera poses tracking is proposed to improve the accuracy of camera tracking and carry out high-fidelity 3D reconstruction.First,a linear-process algorithm is used to establish more reliable dense correspondence between consecutive RGB-D frames.Then,the dense correspondence is used to improve the accuracy of traditional ICP(iterative closest point)method for position and pose tracking of RGB-D cameras.Despite the depth map obtained by hand-held RGB-D cameras contains obvious noise and jitter is easy to occur between adjacent frames,the proposed dense correspondence optimization approach significantly improves the accuracy of the RGB-D camera tracking.Furthermore,the algorithm can be effectively applied in closed-loop detection,and can effectively reduce camera drift caused by camera tracking.The most important is that this method could also serve as a plug-in component to the existing ICP-based methods for improving the precision of the camera tracking.Experimental results show that the proposed method is effective and robust both on public data sets and real scenes scanned in real time.