Probabilistic 3D ICP algorithm based on ORB feature

Aligning the 3D point clouds is considered as a crucial step to build consistent maps from unknown environment in SLAM problem of mobile robotics. However, ICP algorithms for aligning the point clouds usually ignore the valuable visual information contained in the point clouds and only model surface structure from the “model” scan. This paper presents a new ICP algorithm incorporating the visual ORB feature. This approach is based on a probabilistic ICP algorithm that takes into account both scans for RGB images along with the depth information from Kinect sensor. Experiments are carried out on real world scenes and results show that the new approach improves the accuracy and saves time of the registration.