Preliminary Study on a Deep Learning 3D Reconstruction Simulation Method Based on Infrared Image Sequence
CHEN Jichi1，WEI Guohua1，GUO Conglong1，ZHANG Lihe2
1. School of Information and Electronics, Beijing Institute of Technology, Beijing 100081,China；
2. School of Information and Communication Engineering, Dalian University of Technology, Dalian 116086, China
Abstract：The paper proposes a 3D reconstruction simulation method based on infrared image sequence.The method combined with traditional multi-view spatial 3D information recovery and deep learning network, which is specifically divided into two parts: the raw point cloud generation, and point cloud completion. In the first part, the incomplete raw point cloud is generated based on the traditional passive multi-view 3D reconstruction method by inputting the pre-processed infrared image sequence. In the second part, the incomplete point cloud is passed through a joint network structure, performing point cloud completion to a 3D reconstruction. Experiments show that this method can effectively realize the point cloud generation of the infrared image sequence and can effectively restore the missing part of the point cloud. Compared with the existing point cloud completion methods, the test results of some categories performed better in EMD (Earth Mover's Distance) index. With the increase of training data and optimization of training methods, it shows a potential to complete the 3D reconstruction task of infrared image sequences with high quality.
陈寂驰, 魏国华, 郭聪隆, 张立和. 一种基于红外图像序列的深度学习三维重建仿真方法初探[J]. 空天防御, 2020, 3(4): 21-29.
CHEN Jichi, WEI Guohua, GUO Conglong, ZHANG Lihe. Preliminary Study on a Deep Learning 3D Reconstruction Simulation Method Based on Infrared Image Sequence. Air & Space Defense, 2020, 3(4): 21-29.
