基于探測(cè)球的固定式掃描海量點(diǎn)云自動(dòng)定向方法
郭敬平
(寧波市鄞州區(qū)測(cè)繪院,浙江 寧波 315100)
摘要:地面固定式掃描點(diǎn)云首先要將自由坐標(biāo)系的點(diǎn)云納入國(guó)家坐標(biāo)系,而單站掃描的點(diǎn)云數(shù)據(jù)量極大,無(wú)法在可視環(huán)境下進(jìn)行拼接。針對(duì)現(xiàn)有方法對(duì)海量點(diǎn)云拼接的不足,提出一種基于探測(cè)球的固定式掃描海量點(diǎn)云自動(dòng)定向方法,該方法通過(guò)數(shù)據(jù)關(guān)聯(lián)技術(shù)讀取海量點(diǎn)云、建立標(biāo)靶搜索環(huán)、球擬合確定標(biāo)靶候選點(diǎn)、全組合距離匹配法確定同名點(diǎn)及坐標(biāo)轉(zhuǎn)換參數(shù)解算等,完成點(diǎn)云的自動(dòng)定向過(guò)程。通過(guò)實(shí)驗(yàn)驗(yàn)證文中算法的有效性及可行性。
關(guān)鍵詞:海量點(diǎn)云;定向標(biāo)靶;點(diǎn)云絕對(duì)定向;扇形等距離索引;全組合距離匹配
中圖分類號(hào):P237
收稿日期:2014-07-14;修回日期:2015-02-28
作者簡(jiǎn)介:郭敬平(1979-),男,高級(jí)工程師.
An auto registration method for fixed type of scanned HPC based on detecting balls
GUO Jin-ping
(Ningbo Institute of Surveying and Mapping Yinzhou District,Ningbo 315100, China)
Abstract:The point cloud data of fixed terrestrial scanning are used for topographic mapping,of which the data should be transformed from free coordinate system into national coordinate frame.However,due to the huge quantity of point cloud for single-station scanning,the data can not be registered in a visual environment.For the deficiency of existing registration methods for huge point clouds (HPC),an auto method for fixed type of scanned HPC based on detecting balls is put forward.The automatic registration process is implemented by reading point clouds through data association, searching target-searching loops,followed by fitting spheres for candidate targets,extracting corresponding points using distance matching, and calculating the transformation parameters.Experiments demonstrate the effectivity and feasibility of the proposed algorithm.
Key words:huge point cloud;orientation target;absolute orientation for point cloud;concentric equidistant index;distant matching of full combination
隨著三維激光掃描儀硬件性能的提高,3D掃描技術(shù)日趨成熟[1],基于地面的激光掃描系統(tǒng)TLS(Terrestrial Laser Scanning)單站掃描點(diǎn)云數(shù)量越來(lái)越大,GB級(jí)的海量點(diǎn)云使得在現(xiàn)有普通計(jì)算機(jī)難以基于內(nèi)存進(jìn)行點(diǎn)云的定向、降噪、植被過(guò)濾、地面點(diǎn)云獲取[2]、地圖要素提取[3]等處理過(guò)程,成為亟待解決的技術(shù)瓶頸。為了將地面固定式掃描點(diǎn)云應(yīng)用于地形測(cè)繪,首先要將自由坐標(biāo)系的點(diǎn)云納入國(guó)家坐標(biāo)系,但單站掃描的點(diǎn)云數(shù)據(jù)量無(wú)法在可視環(huán)境下用文獻(xiàn)[4]和[5]中方法進(jìn)行兩站或多站點(diǎn)云的拼接。地形測(cè)繪中,兩站掃描的點(diǎn)云僅有一小部分重疊且重疊的位置未知[6],無(wú)法用迭代最近點(diǎn)法ICP(Iterative Closest Point)[7]實(shí)現(xiàn)基于表面特征匹配[8]和區(qū)域形狀分析法[9]的自動(dòng)拼接,需要新方法將點(diǎn)云納入到測(cè)量的坐標(biāo)系中。
為此,本文在充分利用球形定向標(biāo)靶具有幾何定向精度高及易于識(shí)別的特性基礎(chǔ)上,提出一種基于探測(cè)球的固定式掃描海量點(diǎn)云自動(dòng)定向方法,包括三個(gè)環(huán)節(jié):①用數(shù)據(jù)關(guān)聯(lián)技術(shù)讀取海量點(diǎn)云;②用等距離扇形分區(qū)法建立點(diǎn)云空間索引;③用全組合距離匹配法確定匹配點(diǎn)(標(biāo)靶球中心的地理坐標(biāo)與掃描坐標(biāo)對(duì)應(yīng)),實(shí)現(xiàn)標(biāo)靶球海量點(diǎn)云自動(dòng)定向。
1海量點(diǎn)云的空間索引和自動(dòng)定向原理
一站點(diǎn)云分布統(tǒng)計(jì)規(guī)律,0.1D(D為掃描儀測(cè)程,高密度區(qū))點(diǎn)占86%,0.1D~0.3D(中密度區(qū))范圍的點(diǎn)占12.3%,大于0.3D(低密度區(qū))的點(diǎn)占1.7%(見(jiàn)圖1),星點(diǎn)一般在中密度區(qū),這個(gè)區(qū)就是搜索區(qū)。根據(jù)GPS測(cè)得各球形標(biāo)靶的國(guó)家地理坐標(biāo)和水準(zhǔn)測(cè)量的高程,確定標(biāo)靶范圍,用數(shù)據(jù)關(guān)聯(lián)技術(shù)分段讀取坐標(biāo),每段讀取1 000萬(wàn)個(gè)點(diǎn),從每段數(shù)據(jù)過(guò)濾出標(biāo)靶所在范圍的點(diǎn),如圖2所示;取出每個(gè)標(biāo)靶所在環(huán)形點(diǎn)云,如圖3所示。
圖1 原始有效范圍點(diǎn)云
對(duì)點(diǎn)云建立空間索引現(xiàn)有的方法有k-d樹(shù)索引結(jié)構(gòu)[10]、十叉樹(shù)空間索引[11]和球形空間索引等[12]。環(huán)形點(diǎn)云按等距離分區(qū),如圖4所示,建立點(diǎn)云扇形等距平面索引數(shù)據(jù)庫(kù)。該方法與矩形分區(qū)的矩形分區(qū)搜索球時(shí),球上的點(diǎn)最多可能分布在四個(gè)區(qū)域,而扇形結(jié)構(gòu)搜索時(shí),則球上的點(diǎn)最多分布在兩個(gè)區(qū)域中,減少了重復(fù)擬合次數(shù)。
圖4 扇形等距離分區(qū)
扇形等距索引相對(duì)于扇形等角度索引的優(yōu)點(diǎn)是能保證離掃描站不同距離的標(biāo)靶搜索格大小基本一致。
標(biāo)靶點(diǎn)云如圖5所示,按照扇形等距平面索引方法,以一個(gè)點(diǎn)云帶為一個(gè)探測(cè)區(qū),將探測(cè)區(qū)利用對(duì)應(yīng)的中心角分成多個(gè)扇形,以每個(gè)扇形為探測(cè)單元,用球面擬合法逐單元進(jìn)行球探測(cè),從此得到一系列擬合半徑與已知半徑之差在2 cm之內(nèi)且不同圓球度的星點(diǎn),而后將圓球度在85%以上的星點(diǎn)作為候選點(diǎn)。候選點(diǎn)的圓球度
(1)
式中:dist(O,Pk)為球面點(diǎn)到球心球面點(diǎn)的無(wú)符號(hào)距離函數(shù)[15],r0為球的已知半徑,r為擬合球半徑,通常選擇C大于85%的為可信球。
圖5 要探測(cè)的球
1)刪除重復(fù)點(diǎn):在確定候選點(diǎn)后,選擇第一個(gè)球,計(jì)算后面所有球到此球中心距離,刪除距離小于2 cm的重復(fù)球。
3)重采樣和星點(diǎn)坐標(biāo)精化:獲得同名點(diǎn)后,對(duì)星點(diǎn)點(diǎn)云進(jìn)行重采樣并進(jìn)行擬合,見(jiàn)圖6、圖7。進(jìn)而刪除擬合誤差大的點(diǎn)(人工刪除結(jié)合算法),二次擬合星點(diǎn)坐標(biāo)。
圖7 擬合球面2
用星點(diǎn)GPS測(cè)量坐標(biāo)和掃描坐標(biāo),平差計(jì)算三維坐標(biāo)轉(zhuǎn)換參數(shù),組成定向矩陣,將掃描坐標(biāo)轉(zhuǎn)換到國(guó)家地理坐標(biāo),實(shí)現(xiàn)進(jìn)行點(diǎn)云的定向,點(diǎn)云定向矩陣定義為
(2)
式中:ΔX,ΔY,ΔZ為平移參數(shù),其實(shí)質(zhì)是掃描儀中心的地理坐標(biāo),a,b,c是反對(duì)稱矩陣中三個(gè)參數(shù),是縮放因子,一般取1 。其它9個(gè)元素是點(diǎn)云繞三個(gè)坐標(biāo)軸三個(gè)旋轉(zhuǎn)角的函數(shù)(構(gòu)成旋轉(zhuǎn)矩陣)。點(diǎn)云定向6個(gè)自由度SDOF (Six Degrees of Freedom)由Besl,McKay[6]給出。
2實(shí)驗(yàn)及分析
本實(shí)驗(yàn)所用數(shù)據(jù)來(lái)自某礦山實(shí)際掃描多站掃描點(diǎn)云,掃描儀器為Riegl VZ-1000。礦區(qū)3.2 km×2.8 km,共掃描47站,每站設(shè)四個(gè)標(biāo)靶球,每站約1億個(gè)點(diǎn)。以第29站為例,說(shuō)明本方法可行性。將4個(gè)排球作為球形標(biāo)靶如圖8所示,放在經(jīng)過(guò)精平的三角基座上,球心三維坐標(biāo)由GPS和水準(zhǔn)測(cè)量獲得。建立等距離扇形索引后,對(duì)每個(gè)扇形區(qū)點(diǎn)進(jìn)行探測(cè)和球面擬合,擬合半徑與球理論半徑相差不大于1 cm,則認(rèn)為該區(qū)可能有標(biāo)靶球(候選球),共18個(gè)候選球,結(jié)果見(jiàn)表1。
表1 標(biāo)靶候選球 cm
然后計(jì)算候選球的圓球度,圓球度大于85%的可信球6個(gè),見(jiàn)表2,刪除重復(fù)球后,就是4個(gè)星點(diǎn)上的球數(shù)據(jù)見(jiàn)表3。
自動(dòng)探測(cè)的結(jié)果與人工刪除噪聲點(diǎn)后的結(jié)果進(jìn)行球面擬合,結(jié)果如表4所示,Q2903球半徑相差比較大,這是因?yàn)楸砻嬗休^多的噪聲點(diǎn),精化算法有待于改進(jìn)。Q2901、Q2902、Q2904星點(diǎn)處半徑差在1 cm之內(nèi),符合要求。為了測(cè)試軟件自動(dòng)定向效率,用不同大小點(diǎn)云,在不同配置的計(jì)算機(jī)進(jìn)行定向,所用時(shí)間見(jiàn)表5。
表2 可信球 cm
表3 定向球坐標(biāo) cm
表5 運(yùn)行時(shí)間
3結(jié)束語(yǔ)
本文提出的方法能實(shí)現(xiàn)單站點(diǎn)云靶標(biāo)自動(dòng)檢測(cè)與快速定向,將多站掃描點(diǎn)云納入國(guó)家坐標(biāo)系,大大提高了點(diǎn)云的拼接效率,提高點(diǎn)云定向精度,降低其對(duì)點(diǎn)云處理及應(yīng)用的影響。
參考文獻(xiàn):
[1]WANDA M,et al..Processing and interactive editing of huge point clouds from 3D scanners[J].Computers & Graphics 2008,32:204-220.
[2]張會(huì)霞,陳宜金,劉國(guó)波.基于三維激光掃描儀的校園建筑物建模研究[J].測(cè)繪工程,2010,19(1):32-34,38.
[2]赫春曉,李浩.基于三維激光掃描數(shù)據(jù)的三維實(shí)景構(gòu)建[J].測(cè)繪工程,2013,22(1):81-85.
[3]VAN de Giessen,VOS F M,GRIMBERGEN C A,VAN Vliet L J,et al.An efficient and robust algorithm for parallel groupwise registration of bone surfaces[J].Med Image Comput Comput Assist Interv.2012;15(3):164-171.
[4]JIAN Yao,MAURO R R,PIERLUIGI T,et al.Robust surface registration using N-points approximate congruent sets[J].EURASIP Journal on Advances in Signal Processing,2011,(1):1-27.
[5]TORRE-FERRERO C,LLATA J R,ALONSO L,et al.3D point cloud registration based on a purpose-designed similarity measure[J].EURASIP Journal on Advances in Signal Processing 2012,(3):1-15.
[6]Ondˇrej Jeˇz.3D Mapping and Localization Using Leveled Map Accelerated ICP[C].// European Robotics Symposium 2008,2008,2:466-473.
[7]LI N,CHENG P,SUTTON M A,et al.Three-dimensional Point Cloud Registration by Matching Surface Features with Relaxation Labeling Method[J].Experimental mechanics,2005,45(1):71-82
[8]HUANG D,et al.3D Face Recognition Based on Local Shape Patterns and Sparse Representation Classifier[J].Advances in Multimedia Modeling,2011,65(23):206-216
[9]JOCHEM A,et al.Area-wide roof plane segmentation in airborne LiDAR point clouds[J].Computers,Environment and Urban Systems,2012,(36):54-64.
[10]Jagan Sankaranarayanan,SAMET H,VARSHNEY A.A fast all nearest neighbor algorithm for applications involving large point-clouds[J].Computers & Graphics,2007,(31):157-174
[12]MA Hongchao,WANGA Zongyue.Distributed data organization and parallel data retrieval methods for huge laser scanner point clouds[J].Computers & Geosciences,2011,(37):193-201. M,et al..Processing and interactive editing of huge point clouds from 3D scanners[J].Computers & Graphics 2008,32:204-220.
[2]張會(huì)霞,陳宜金,劉國(guó)波.基于三維激光掃描儀的校園建筑物建模研究[J].測(cè)繪工程,2010,19(1):32-34,38.
[2]赫春曉,李浩.基于三維激光掃描數(shù)據(jù)的三維實(shí)景構(gòu)建[J].測(cè)繪工程,2013,22(1):81-85.
[3]VAN de Giessen,VOS F M,GRIMBERGEN C A,VAN Vliet L J,et al.An efficient and robust algorithm for parallel groupwise registration of bone surfaces[J].Med Image Comput Comput Assist Interv.2012;15(3):164-171.
[4]JIAN Yao,MAURO R R,PIERLUIGI T,et al.Robust surface registration using N-points approximate congruent sets[J].EURASIP Journal on Advances in Signal Processing,2011,(1):1-27.
[5]TORRE-FERRERO C,LLATA J R,ALONSO L,et al.3D point cloud registration based on a purpose-designed similarity measure[J].EURASIP Journal on Advances in Signal Processing 2012,(3):1-15.
[6]Ondˇrej Jeˇz.3D Mapping and Localization Using Leveled Map Accelerated ICP[C].// European Robotics Symposium 2008,2008,2:466-473.
[7]LI N,CHENG P,SUTTON M A,et al.Three-dimensional Point Cloud Registration by Matching Surface Features with Relaxation Labeling Method[J].Experimental mechanics,2005,45(1):71-82
[8]HUANG D,et al.3D Face Recognition Based on Local Shape Patterns and Sparse Representation Classifier[J].Advances in Multimedia Modeling,2011,65(23):206-216
[9]JOCHEM A,et al.Area-wide roof plane segmentation in airborne LiDAR point clouds[J].Computers,Environment and Urban Systems,2012,(36):54-64.
[10]Jagan Sankaranarayanan,SAMET H,VARSHNEY A.A fast all nearest neighbor algorithm for applications involving large point-clouds[J].Computers & Graphics,2007,(31):157-174
[12]MA Hongchao,WANGA Zongyue.Distributed data organization and parallel data retrieval methods for huge laser scanner point clouds[J].Computers & Geosciences,2011,(37):193-201.
[責(zé)任編輯:李銘娜]