WEmap: Weakness-Enhancement Mapping for 3D Reconstruction with Sparse Image Sequences

Kun Zhang, Chunying Song, Jingzhao Wang, Kai Wang, Nan Yun

Abstract: Previous studies assume that a dense image sequence can be used for 3D reconstruction because the images are easily captured by mobile devices. However, mobile devices are not applicable in some cases, such as smart factories, which require real-time monitoring and site safety. Therefore, conducting 3D reconstruction with sparse image sequences is important to reduce the number of used devices, and thus, lower the cost of image acquisition. In this study, we propose weakness-enhancement mapping (WEmap) to improve the results of 3D reconstruction based on sparse image sequences. After the initial reconstruction, the contribution of each image is evaluated by mapping the 3D point cloud to 2D images. The low-contribution images and corresponding matching images are weighted to enhance the weaknesses of the initial reconstruction. To the best of our knowledge, this is the first study on 3D reconstruction with sparse image sequences. Experimental results on the sparse DTU [1] and sparse Tanks & Temples [3] datasets demonstrate that WEmap can effectively enhance a reconstructed structure.