基于标记点丢失的多幅自标定图像的3维重建和相机姿态恢复

Reconstruction and Camera Poses Recovery from Multi Self-calibration Images with Marked Point Occlusions

为了对标记点丢失的多幅自标定图像进行精确重建,提出了一种基于标记点丢失的多幅自标定图像的3维重建和相机姿态恢复的方法。该方法与原来方法的不同之处在于,该方法是利用标记点(编码点和非编码点)的方式,即用编码点进行单CCD相机的自标定和姿态恢复,而用非编码点进行3维点的3维重建。该方法有以下3个主要特点:(1)由于该方法采用了标记点的自动识别匹配,所以避免了手工交互选择图像点对(point correspon-dences)费工费时的问题;(2)由于标记点匹配精确,提高了3维点的重建精度,故符合工程要求;(3)由于噪音对标记点的像点影响较小,因此该方法比以前的方法具有更好的鲁棒性。实验结果表明,利用该方法产生的3维重建点精确可靠,能够满足逆向工程等应用的要求。

In order to reconstruct accurately for multi self-calibratlon images with marked point occlusions, this paper proposes a method for reconstruction and camera poses recovery from multi self-calibration images with marked point occlusions. Compared to previous methods, this method uses a way of marked point： coded point and non-coded point. The coded points are utilized for a single CCD camera self-calibration and camera poses recovery and the non-coded points are utilized for 3D points reconstruction. The novelties of the proposed method are three-fold. Firstly, the problem of manually selected image point correspondences is avoided by using auto match of marked points. Secondly, the proposed method is based on marked points, the precision of 3D reconstruction is improved since match of marked point is accurate. Thirdly, the robustness is markedly increased since the effect of noise on the marked points of images is small. The practice shows that the results are accurate and reliable so as to meet the requirements of reverse engineering applications.