Dynamic scenes reconstruction based on foreground and background splitting

Abstract

In recent years, several approaches have been proposed to reconstruct static scenes. However, there exists rarely method for dynamic scenes. In this paper, we propose an approach to reconstruct dynamic scenes using only a moving depth sensor. The main idea is to reconstruct the moving foreground objects and static background scenes respectively. An iterative closest point (ICP) algorithm is used for obtaining the current camera pose. The foreground and background splitting is achieved by processing failed tracking points. We introduce an efficient filter and morphological operations to handle these points. Finally, foreground and background models are reconstructed separately using the split depth data and the calculated camera pose. Experimental results demonstrate that our method has fast and robust performance in reconstructing complex dynamic scenes.