6D-slam with navigable space discovering

To create global map we take obstacles as natural landmarks, it is however necessary to find the correspondences between landmarks by use of a Bhatacharayya distance. Our experiments demonstrate the functionality of estimating the exact pose of the robot and the consistence of the global map of the environment.
Nội dung trích xuất từ tài liệu:
6D-slam with navigable space discoveringJournal of Automation and Control Engineering, Vol. 1, No. 2, June 20136D-Slam with Navigable Space DiscoveringBoutine RachidDepartment of computer science 20 august 1955 university, Skikda, AlgeriaBoutine_rachid@yahoo.frBenmohammed MohamedLIRE Laboratory Mentouri university, Constantine, AlgeriaBen_moh123@yahoo.comAbstract—For all mobiles robots that navigate in unknownoutdoor environments two basic tasks are required:discovering navigable space, and obstacles detection. In thiswork, we have proposed a new variant of the 6d-slamframework, wherein planar patches were used forrepresenting navigable area, and 3d mesh for representingobstacles. A decomposition of the environment in smallervoxels is made by an octree structure. Adjacent voxels withno empty intersection of their best fitting plans should bepiece together in the same navigable space; otherwiseadjacent non planar voxels would form the seed of potentialobstacles. To create global map we take obstacles as naturallandmarks, it is however necessary to find thecorrespondences between landmarks by use of aBhatacharayya distance. Our experiments demonstrate thefunctionality of estimating the exact pose of the robot andthe consistence of the global map of the environment. Index Terms—6d-Slam,operation, OctreeI.Bhatacharayyadistance,the fourth section, we present a solution for navigablespace growth problem, and in the fifth section, wepropose a novel association method, in the sixth section,we present the proposed 6d-slam variant, finally in thelast section, we give the experimental results.II.In the last decade, a great deal of latest works areoriented towered 6D-Slam, like the work of Paloma de laPuente and all [1], [2], where they present a segmentationand fitting algorithm of 3d points clouds, based oncomputer vision techniques, wherein a least-squaresfitting, and a maximum Incremental probability algorithmformulated upon the Extended Kalman Filter, were usedto outcome a position in 6D, and a map of planar patcheswith a convex-hull.In literature there exist two kinds of map: The firstkind is topological maps, which consists of a set of scansconnected by a network of edges in a tree fashion style,like the work of D. Borrman [3], where a graphslammethod is presented, a sparse network to represent therelation between several overlapping 3D scans. Thesecond kind is metric maps, which are more reliable torepresent geometrical features of the environment, eitherfree, or occupied space.More specifically in metric maps there are two mainapproaches: firstly, using feature extracted map, whichallows the reduction of the amount of data stored in robotmemory, nevertheless it suffers from information loss,secondly, the use 3D raw data directly, although the hugestorage space required like the work of R. Schnabel in [4].Elevation map was proposed by W. Miao and all, [5] torepresent the environment as digital elevation. Also, theparticle filter on multilevel space was presented by M, V.Prieto, and all in [6], or the work of K. Rainer, which iscited in [7].Furthermore, M, P. Imtyas have present a newrepresenting model called GP (Gaussian process model),where the robot uses this model to localize whiletraversing each environment [8]. On the other hand,textured mesh is proposed by C. Beall in [9] where atechnique for large scale sparse reconstruction ofunderwater structures is presented; this techniqueestimates the trajectory of camera, and 3D feature poses.DotINTRODUCTIONMajor efforts and publications of robotic community inlast decade are oriented toward the development of newgeneration of robots able to navigate in outdoorenvironment, although the GPS signal is poor, or missing.The apparition of new kinds of 3d sensors as 3d lasertelemeters, time of flight, and kinect cameras; havecontribute in the development of more sophisticatedrobotic applications.Our contribution in this work is the proposition of newvariant of 6d-slam framework, wherein, we split the taskof mapping in two sub tasks: obstacles detection, andnavigable space discovering.We will demonstrate that this separation contributes inthe outcome of other robotic sub tasks like: localization,path planning, and obstacle avoidance.Lot of domains can benefit from the results of thiswork, such as mapping underground mines, industrialautomation, unmanned transportation, disaster rescuemission, etc.This paper is organized as follow: in first section westart by an introduction, in the second section, we make asurvey on the most important related works, in the thirdsection, we present a new obstacle detection algorithm, inManuscript received October 20, 2012; revised December 25, 2012.©2013 Engineering and Technology Publishingdoi: 10.12720/joace.1.2.78-81RELATED WORK78Journal of Automation and Control Engineering, Vol. 1, No. 2, June 2013An alternative way was the split of the 3d raw data, inclusters of low resolution, such as voxel that divide thespace in cubic units, like the work of T. Suzuki and othersin [10]. Other researchers have proposed a recursivespace decomposition method, similar to tree basedrepresentation, like the work of N. Fairfield, and allwhere an octree is used [11], to represent 3d environment,therefore a well distinction between occupied, free, andunknown area is possible.In our proposed method, we consider a navigable spacefor a mobile robot as a set of contiguous coplanar patches.A navigable space is created by merging a subset ofadjacent planar leaf voxels, on which there exists at leastan intersection boundary’s line of their best fitting plans,according to the following algorithm:Algorithm 1: NSD(octree)Inputs: Octree of the scanned environmentFor all childrenIf (children is leaf node) then { NSD(children) }Else{(p1, p2, p3, p4) =find the intersect ...