3D map repair after moving objects removing

Abstract

Reconstructions of outdoor and indoor environments using stereo cameras are confronted with harmful effects caused by moving objects. Ghosting shown in Fig.1 not only leads to misunderstanding but also hamper the reconstruction of the static scene. 3D map with only still objects can be used for robotics navigation after moving-object detection. Nevertheless, blanks maybe occurred in the 3D map once removing the moving object. In this paper, we detect the moving objects among consecutive stereo image pairs and reconstruct with the static objects only. Furthermore, we are the first to introduce image inpainting methods into 3D reconstruction and using plane smoothing method to fill holes left by moving objects in disparity images. We evaluate our results on challenging scenarios, and the results show our algorithm can achieve good performances in reconstruction dealing with moving objects. Besides, comparison of reconstruction results among different inpainting methods is demonstrated.