Dense Color Constraints based 6D object pose estimation from RGB-D images

A key problem for 6D pose estimation based on RGB-D image input is how to make full use of these two different data sources. The previous work simply took the depth map as the input of the fourth channel of CNN, or carried out the fusion offeatures extracted from these two data sources with different methods. But their fusion did not impose the right constraints and lost some valuable information. In this work, we propose that DCC (Dense Color Constraints). 6D pose estimation performance can be improved effectively by using dense corresponding color constraints. Experiments show the most advanced end-to-end performance in LineMod datasets.