Research On Fast And Robust Algorithm Of The Map Point Set Registration

Map point set registration is a key technology in the field of vehicle navigation,mobile robot, and panoramic map. It makes the space position of two different setsuniform after space transformations. As the development of computer vision, imageprocessing and network technology, the algorithms need to meet the real-timerequirements in practical applications, so it is necessary to research on the fast androbust algorithms. Based on this, this paper forcuses on the fast and robust algorithmsfor the map point sets registration. The main work is as follows:Firstly, map point sets have some features such as more points, complex structure,and are easily affected by the sensor noise, thus the registration has low precision andtime-consuming, it can not meet the real-time requirements of autonomous driving.This paper proposes a multi-scale ICP algorithm （Multi-scale Iterative ClosestPoints, MSICP）using the thought of hierarchy based on the classic ICP（IterativeClosest Points）algorithm to improve the speed and accuracy. MSICP firstly makesthe original point sets multi-scaled; then uses the translation of multi-scale point setsthat have been matched as the initial translation of the originals to complete the quickand precise registration. Experimental results show that the improved algorithmperforms better in the speed and precision and has some theory and practical values.Secondly, on the basal research of MSICP for rigid point sets, we propose a multi-scale scale ICP algorithm (Multi-scale Scale Iterative Closest Points, MSSICP) fornon-rigid map point sets registration. MSSICP firstly makes the original non-rigidpoint sets multi-scaled and matches the sparse point sets; then uses the translation ofthe matched non-rigid sparse point sets as the initial translation of originals; at last itcompletes the fast and robust registration for the isotropic and anisotropic scaledpoint sets. Experimental results prove the MSSICP algorithm also performs better.At last, as map point sets have more points, complex structure and are easilyinterfered by the noise but robust, we proposed a improved ICP algorithm based onskeleton (Skeleton Iterative Closest Point, SKICP), which uses skeleton algorithmgenerate sparse representation of map structure to improve the speed and accuracy.SKICP firstly extracts the skeleton of sparse point sets, then uses the translation ofthe matched skeletons of the sparse sets as the initial translation of the originals tocomplete the quick and precise registration. Experimental results show the SKICPalgorithm performs better and has the significance of theory and reference.