基于图像分割的三维点云深度值合成

3D point cloud depth synthesis based on image segmentation

针对传统的计算机视觉方法对复杂物体重建效果不完整的缺点,提出一种基于图像分割的三维点云深度值合成算法。该算法将输入图像过分割为一系列形状、大小相近的超像素,用图结构的方法找到与重建效果欠佳的目标超像素颜色、距离均相近的源超像素,并将源超像素的深度信息传播到目标超像素区域。实验证明,该算法能够修补三维点云中缺失的深度,最终改善三维点云的重建效果。

Aiming at the poor performance of using traditional computer vision based methods to reconstruct complicated objects,this paper proposed a depth synthesis algorithm based on segmentation of input images. The algorithm over-segmented input images into a set of superpixels in nearly the same shape and size,and found source superpixels both similar and approximate to the poor reconstructed target superpixels using graph structure. Then,it propagated the depth information of source superpixels to the target superpixels. Experiments have proved that the method can integrate the missing depth information of 3D point cloud and improve the 3D reconstruction result.