|Table of Contents|

Nature Terrain Classification Using Point Cloud Data


Research Field:
Publishing date:


Nature Terrain Classification Using Point Cloud Data
YUAN XiaZHAO Chun-xiaZHANG Hao-fengCAI Yun-fei
School of Computer Science and Technology,NUST,Nanjing 210094,China
robot navigation terrain classification point clouds multi-feature Gaussian mixture model
A nature terrain classification algorithm is proposed to deal with the environment understanding problem in moving robot navigation.Based on the single geometrical feature terrain classification method,a complex feature extracted from point cloud is used to train a classifier.The complex feature vector includes both geometrical feature and color feature of a point.The algorithm computes the eigenvalue of the coordinate covariance matrix and the average normal vector of a point as a geometrical feature.The points obtain color information by matching coordinates of Lidar and camera.The color of points is added into complex feature vector as a color feature.An expectation-maximization Gaussian mixture model(EM-GMM) is employed to train a classifier.The training data are labeled by man.The experimental results show that: compared with the single geometric feature classification method,the complex feature classification method obtains higher correctness in classifying nature terrain.


[1]H uang J, Lee A, M um fo rd D. Statistics o f range im ages[ A ]. Proceedings o f the ComputerV ision and Pa-t te rn Recognition [ C ]. Lo s A lam itos, CA, Un ited States: IEEE, 2000: 1324- 1331.
[2] M acedo J, M anduchiR, M atth ies L. Ladar-based d iscr im ina tion o f grass from obstac le fo r autonomous nav-i gation[ A] . Proceedings o f Interna tiona l Sym pos ium on Experim enta l Robotics [ C ]. London, UK: Springer- Ver lag, 2000: 111- 120.
[3] Castano A, M a tthies L. Fo liage d isc im ina tion us ing a rota ting lada r[ A ]. Proceed ings of IEEE Inte rnational Conference on Robotics and Autom ation [ C ]. Lo s A lam itos, CA, United Sta tes: IEEE, 2003: 1- 6.
[4] W e llington C, Stentz A. Learn ing pred iction o f the load-bearing surface fo r autonomous rough-terra in navigation in veg etation[ A ]. Inte rnational Confe rence on Fie ld and Se rv ice Robo tics[ C ]. Los A lam itos, CA, Un ited States: IEEE, 2003: 49- 54.
[5] VandapelN, H uber D, KapuriaA, et a.l Natura l terrain c lassification using 3-D ladar data [ A ]. IEEE International Con ference on Robotics and Automation[ C]. Los A lam itos, CA, Un ited States: IEEE, 2004: 5117- 5122.
[6] S itho le G, Vosselm an G. ISPRS com pa rison of filters [ R]. Delft, Nethe rland: Netherlands Comm ission III W o rking Group 3, De lft Univers ity o f Techno logy, 2003: 1- 29.
[7] 韩光, 赵春霞. 融合多可视化特征的可通行性地形 分类[ J]. 华中科技大学学报( 自然科学版), 2008, 36( SI): 105- 108.
[8] 李旭涛, 彭复员, 曹汉强, 等. 地形表面的自相似程 度与分类感知[ J] . 电子与信息学报, 2007, 29( 6): 1480- 1482.
[9] 王琤, 胡鹏, 刘晓航, 等. 基于数字地形分析的火星 地貌自动化分类方法[ J]. 武汉大学学报(信息科 学版) , 2009, 34( 4): 483- 487.
[10] Vandape lN, HuberD, Kapur iaA, et a.l Natura l terrain classifica tion using 3-D ladar data [ A ]. IEEE Internationa lConference on Robo tics and Automation[ C ]. Los A lam itos, CA, Un ited States: IEEE, 2004: 5117- 5122.
[11] Yuan X ia, Guo Ling, W ang Jian-yu, et a.l E ffic ient K-nearest ne ighbors searching a lgor ithm s for unorganized cloud po ints[ A]. The 7thW or ld Cong ress on Intelligen t Control and Autom ation [ C ]. Chongq ing, Ch ina: Institute o f E lectrical and E lectron ics Eng ineers Inc, 2008: 8507- 8510.
[12] 刘大学, 戴斌, 李政, 等. 一种单线激光雷达和可见 光摄像机的标定方法[ J]. 华中科技大学学报( 自 然科学版), 2008, 36( SI): 68- 71.


Last Update: 2010-04-30