RESUMO
Photon-efficient 3D reconstruction under sparse photon conditions remains challenges. Especially for scene edge locations, the light scattering results in a weaker echo signal than non-edge locations. Depth images can be viewed as smooth regions stitched together by edge segmentation, yet none of the existing methods focus on how to improve the accuracy of edge reconstruction when performing 3D reconstruction. Moreover, the impact of edge reconstruction to overall depth reconstruction hasn't been investigated. In this paper, we explore how to improve the edge reconstruction accuracy from various aspects such as improving the network structure, employing hybrid loss functions and taking advantages of the non-local correlation of SPAD measurements. Meanwhile, we investigate the correlation between the edge reconstruction accuracy and the reconstruction accuracy of overall depth based on quantitative metrics. The experimental results show that the proposed method achieves superior performance in both edge reconstruction and overall depth reconstruction compared with other state-of-the-art methods. Besides, it proves that the improvement of edge reconstruction accuracy promotes the reconstruction accuracy of depth map.