郭彩玲,宗 澤,張 雪,劉 剛
?
基于三維點(diǎn)云數(shù)據(jù)的蘋(píng)果樹(shù)冠層幾何參數(shù)獲取
郭彩玲1,2,宗 澤1,張 雪1,劉 剛1※
(1. 現(xiàn)代精細(xì)農(nóng)業(yè)系統(tǒng)集成研究教育部重點(diǎn)實(shí)驗(yàn)室,農(nóng)業(yè)部農(nóng)業(yè)信息獲取技術(shù)重點(diǎn)實(shí)驗(yàn)室,中國(guó)農(nóng)業(yè)大學(xué),北京100083;2. 唐山學(xué)院機(jī)電工程系,唐山063000)
針對(duì)果園環(huán)境下蘋(píng)果樹(shù)冠層參數(shù)獲取精度較低的問(wèn)題,提出了基于地面三維激光掃描儀高精度獲取蘋(píng)果樹(shù)冠層參數(shù)的方法。選用Trimble TX8地面三維激光掃描儀作為蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)采集設(shè)備,提出了基于標(biāo)靶球的KD-trees-ICP算法,用于高精度配準(zhǔn)蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)。研究了平均風(fēng)速小于4.5 m/s時(shí),距離地面三維激光掃描儀不同遠(yuǎn)近條件下的標(biāo)靶球配準(zhǔn)殘差和擬合誤差的變化規(guī)律,分析結(jié)果表明,標(biāo)靶球平均配準(zhǔn)殘差為1.3 mm,平均擬合誤差為0.95 mm,低于大場(chǎng)景測(cè)量配準(zhǔn)誤差要求(5 mm)。為了提高有風(fēng)環(huán)境下提取蘋(píng)果樹(shù)冠層參數(shù)的精度,研究了0.9~4.5 m/s區(qū)間平均風(fēng)速影響下的蘋(píng)果樹(shù)冠層枝干、果實(shí)、葉片的三維點(diǎn)云質(zhì)量,建立了風(fēng)速與葉片側(cè)面厚度的曲線擬合模型,分析結(jié)果表明,在果園平均風(fēng)速小于1.6 m/s時(shí)可以從蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)中提取高精度冠層參數(shù)。利用地面激光三維掃描儀獲取距離蘋(píng)果樹(shù)12 000 mm以內(nèi)冠層參數(shù),測(cè)量精度高于人工測(cè)量,相對(duì)誤差小于4%,為果樹(shù)高通量信息獲取提供了技術(shù)支持。
測(cè)量誤差;激光掃描儀;精度;蘋(píng)果樹(shù)冠層;三維點(diǎn)云數(shù)據(jù)
果樹(shù)冠層形狀、結(jié)構(gòu)、體積[1]和葉面積指數(shù)、生物量估算[2]與植物生物力學(xué)建模[3]有密切的關(guān)系,是研究果樹(shù)表型的基礎(chǔ),同時(shí)精準(zhǔn)測(cè)量冠層結(jié)構(gòu)和形態(tài)參數(shù)也為研究果樹(shù)精準(zhǔn)噴藥[4]、冠層光照分布[5-6]、果品質(zhì)量研究[7]和自動(dòng)化收獲[8]等提供技術(shù)支持。傳統(tǒng)的手工測(cè)量方法依靠人工經(jīng)驗(yàn),獲取實(shí)地?cái)?shù)據(jù),再將調(diào)查結(jié)果外推到管理尺度作為依據(jù)加以應(yīng)用[9],這種測(cè)量和研究方法在過(guò)去相當(dāng)長(zhǎng)的時(shí)間內(nèi)發(fā)揮了較大作用。隨著精細(xì)農(nóng)業(yè)的發(fā)展,傳統(tǒng)的手工測(cè)量方法在果樹(shù)冠層參數(shù)快速精確測(cè)量方面逐漸呈現(xiàn)出局限性。
激光測(cè)量技術(shù)在林木冠層參數(shù)測(cè)量中的應(yīng)用,尤其用于樹(shù)木和作物的無(wú)損檢測(cè),和傳統(tǒng)的伐倒檢測(cè)、人工測(cè)量方法相比較,較大的減少了木材的損耗和人力物力的消耗,提高了測(cè)量速度。激光測(cè)量技術(shù)在作物冠層參數(shù)測(cè)量中的應(yīng)用主要有,一是大尺度獲取作物冠層三維信息,提取諸如森林樹(shù)木冠層高度[10],籬壁式蘋(píng)果園蘋(píng)果樹(shù)株高[11-12]、冠層三維結(jié)構(gòu)形態(tài)[13]、冠層三維輪廓和樹(shù)冠體積[14]、群體冠層結(jié)構(gòu)[15]、樹(shù)木枝干形態(tài)[16]等,這些研究都是在大尺度上,對(duì)冠層三維結(jié)構(gòu)進(jìn)行的研究。二是針對(duì)果樹(shù)冠層內(nèi)部器官形態(tài)的研究,如花序三維形態(tài)結(jié)構(gòu)[17]和植物葉片三維形態(tài)[18-21],這些均在實(shí)驗(yàn)室環(huán)境中植物器官形態(tài)參數(shù),取得了一定的研究成果。但在果園自然條件下精準(zhǔn)獲取果樹(shù)冠層參數(shù)方法還需要進(jìn)一步研究。
本研究以果園環(huán)境下生長(zhǎng)的蘋(píng)果樹(shù)冠層作為研究對(duì)象,選用地面三維激光掃描儀快速獲取果樹(shù)冠層三維點(diǎn)云數(shù)據(jù),提出一種準(zhǔn)確配準(zhǔn)三維點(diǎn)云的方法,定量分析不同風(fēng)速影響下蘋(píng)果樹(shù)冠層三維點(diǎn)云質(zhì)量,提取蘋(píng)果樹(shù)冠層高精度參數(shù),與人工測(cè)量結(jié)果進(jìn)行對(duì)比分析。研究結(jié)果可為枝葉間異速生長(zhǎng)檢測(cè)、冠層光照時(shí)空分布等提供技術(shù)支持。
1.1 試驗(yàn)方法
由于被掃描目標(biāo)之間相互遮擋,在地面三維激光掃描儀獲取三維數(shù)據(jù)時(shí),只根據(jù)一站點(diǎn)云無(wú)法獲取目標(biāo)完整的三維信息,因此掃描時(shí)需要地面三維激光掃描儀圍繞掃描目標(biāo)物進(jìn)行多站掃描,獲取目標(biāo)物不同部位的三維點(diǎn)云數(shù)據(jù)。
圖1描述了獲取單株蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)的試驗(yàn)方法。地面三維激光掃描儀位于距離最近樹(shù)干中心3.5~5 m的區(qū)域,圖中采用了5站式掃描。為了能夠?qū)?站掃描得到的三維點(diǎn)云數(shù)據(jù)配準(zhǔn)在一起,采用標(biāo)直徑為100 mm靶球作為配準(zhǔn)基準(zhǔn)。針對(duì)多站數(shù)據(jù)配準(zhǔn)時(shí)出現(xiàn)累計(jì)誤差問(wèn)題[22],考慮到果樹(shù)冠層的尺寸,將標(biāo)靶球布置于蘋(píng)果樹(shù)冠層周圍,距地面三維激光掃描儀1~12 m區(qū)域內(nèi),保證每站均可以掃描到4個(gè)以上標(biāo)靶球,并且至少3個(gè)標(biāo)靶球要落在掃描的重疊區(qū)域,利用KD-tree- ICP和標(biāo)靶球輔助配準(zhǔn)的方法配準(zhǔn)點(diǎn)云,并提取冠層參數(shù),通過(guò)定量分析不同距離標(biāo)靶球的擬合誤差、不同風(fēng)速下葉片點(diǎn)云的分層情況,提出果園環(huán)境下蘋(píng)果樹(shù)冠層參數(shù)精準(zhǔn)獲取的方法。
1.2 儀器與材料
地面三維激光掃描儀按測(cè)距原理可分為:基于脈沖型、基于相位差型和基于光學(xué)三角測(cè)量型3種。前2種適用于遠(yuǎn)距離測(cè)量,后者可以近距離獲取比較精確的三維點(diǎn)云數(shù)據(jù)。
三維點(diǎn)云數(shù)據(jù)通常包含有掃描物體的三維坐標(biāo)信息、強(qiáng)度信息,部分會(huì)帶有RGB顏色信息(例如trimble TX5)。為了實(shí)現(xiàn)無(wú)損、快速獲取蘋(píng)果樹(shù)冠層的三維點(diǎn)云數(shù)據(jù),試驗(yàn)采用美國(guó)Trimble公司地面三維激光掃描儀TX8進(jìn)行蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)采集(見(jiàn)圖2)。最大掃描范圍為340 m,測(cè)量速度為1 000 000點(diǎn)/s,視場(chǎng)角為360o×317o,精度為0.5",掃描范圍為±10¢,采用脈沖激光測(cè)距,測(cè)量精度不小于0.5″。100 m測(cè)距時(shí),誤差≤2 mm。
a. 三維激光掃描儀Trimble TX8 b. 戴維斯氣象站DAVIS6162
在已建立的中國(guó)農(nóng)業(yè)大學(xué)蘋(píng)果樹(shù)采摘機(jī)器人試驗(yàn)基地(北京市昌平區(qū)南口鎮(zhèn)辛力莊村)內(nèi),開(kāi)展基于地面三維激光掃描儀的蘋(píng)果樹(shù)冠層參數(shù)獲取研究工作。蘋(píng)果樹(shù)株距2.5 m,行距5 m,樹(shù)高3.2~5.1 m,行間生草,灌水條件良好,采用常規(guī)管理方式進(jìn)行春季修剪。本文數(shù)據(jù)采集對(duì)象為隨機(jī)選擇果園自然生長(zhǎng)狀態(tài)下樹(shù)齡7 a的自由紡錘形宮藤富士蘋(píng)果樹(shù)。利用地面三維激光掃描儀獲取蘋(píng)果樹(shù)不同風(fēng)速、不同生長(zhǎng)階段的三維點(diǎn)云數(shù)據(jù)。2015年5月—2016年5月期間,對(duì)果園蘋(píng)果樹(shù)冠層進(jìn)行了7次數(shù)據(jù)采集,初步分析可知,果園現(xiàn)場(chǎng)環(huán)境影響三維點(diǎn)云質(zhì)量。本試驗(yàn)不考慮掃描環(huán)境諸如溫濕度、大氣壓、光照度等參數(shù),僅分析風(fēng)速對(duì)影響掃描參數(shù)的影響。風(fēng)速數(shù)據(jù)由距離掃描目標(biāo)樹(shù)木距離25 m的DAVIS 6162提供。
1.3 點(diǎn)云配準(zhǔn)方法
為了保證點(diǎn)云的配準(zhǔn)精度,按照?qǐng)D1試驗(yàn)方法,掃描之前進(jìn)行儀器調(diào)平,即儀器垂直于地面放置。獲取的蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)如圖3所示,圖3c中綠色框表示掃描目標(biāo)蘋(píng)果樹(shù)冠層,黃色三角是地面三維激光掃描儀的工作位置,稱之為站,每站獲取的三維點(diǎn)云數(shù)據(jù)如圖3 a、b、d、e、f所示,圖3 c是配準(zhǔn)后的三維點(diǎn)云數(shù)據(jù)。
圖3 三維點(diǎn)云數(shù)據(jù)配準(zhǔn)過(guò)程
不同站的點(diǎn)云轉(zhuǎn)換到同一坐標(biāo)系下,稱之為三維點(diǎn)云的配準(zhǔn)。ICP(iterated closest points)算法[23-24]是配準(zhǔn)精度較高的算法,適合于配準(zhǔn)剛體的點(diǎn)云數(shù)據(jù),通過(guò)搜索緊鄰點(diǎn)間的對(duì)應(yīng)關(guān)系,利用坐標(biāo)變換可實(shí)現(xiàn)坐標(biāo)匹配工作。利用ICP算法配準(zhǔn)多站三維點(diǎn)云數(shù)據(jù)時(shí),耗時(shí)較長(zhǎng)。近幾年的研究中,科研人員根據(jù)不同適用場(chǎng)合的三維點(diǎn)云數(shù)據(jù),研究了ICP的改進(jìn)算法[25-27],試圖提高配準(zhǔn)效率。本文進(jìn)行多站三維點(diǎn)云數(shù)據(jù)配準(zhǔn)時(shí)不考慮掃描形變,即認(rèn)為掃描中蘋(píng)果樹(shù)冠層各個(gè)器官為剛體,掃描過(guò)程中形態(tài)不發(fā)生變化。提出KD-tree-ICP算法[28]結(jié)合標(biāo)靶球配準(zhǔn)的方法,提高三維點(diǎn)云數(shù)據(jù)配準(zhǔn)精度。算法流程圖如圖4所示。
圖4 算法流程圖
在提取標(biāo)靶球點(diǎn)云時(shí),需要在圖3每站點(diǎn)云數(shù)據(jù)中,選出標(biāo)靶球點(diǎn)云并手工去除異常點(diǎn),提高標(biāo)靶球球心的定位精度。任選一站三維點(diǎn)云數(shù)據(jù)作為參考點(diǎn)云,例如選取第3站三維點(diǎn)云數(shù)據(jù)做為參考點(diǎn)云(圖3中記為0),其他站三維點(diǎn)云數(shù)據(jù)為源點(diǎn)云。設(shè)掃描得到標(biāo)靶球點(diǎn)云坐標(biāo)為(,球心坐標(biāo)為(),球面半徑為0,按照公式(1)計(jì)算標(biāo)靶球球心,
式中為旋轉(zhuǎn)矩陣,分別表示沿著軸的旋轉(zhuǎn)角,為平移矩陣,表示沿著軸的位移量。
以標(biāo)靶球球心坐標(biāo)為基礎(chǔ),用KD-Tree在各站中搜索同名標(biāo)靶球,利用公式(2)將各站源點(diǎn)云轉(zhuǎn)換到目標(biāo)點(diǎn)云坐標(biāo)下。為了減小累計(jì)誤差,配準(zhǔn)時(shí),每站標(biāo)靶球單獨(dú)參與配準(zhǔn)。即,源點(diǎn)云1和參考點(diǎn)云0配準(zhǔn),得到01,然后源點(diǎn)云1與配準(zhǔn)好的點(diǎn)云01配準(zhǔn),得到02,以此類推,直到所有的源點(diǎn)云配準(zhǔn)完畢。
1.4 點(diǎn)云配準(zhǔn)分析
為了分析蘋(píng)果樹(shù)冠層掃描點(diǎn)云數(shù)據(jù)的質(zhì)量,考慮到樹(shù)干和果實(shí)的不規(guī)則性,利用誤差函數(shù)加速尋找同名點(diǎn)的最小歐氏距離,作為標(biāo)靶球配準(zhǔn)殘差[29],如圖5所示。標(biāo)靶球擬合誤差是表征枝干和果實(shí)的配準(zhǔn)質(zhì)量的另一個(gè)重要參數(shù),表示掃描獲得的目標(biāo)物三維點(diǎn)云的形變程度。按照隨機(jī)抽樣一致法RANSAC(RANdom Sample Consensus)[30],擬合多站標(biāo)靶球三維點(diǎn)云數(shù)據(jù)后的擬合誤差。可以看出,在果園自然環(huán)境下,平均風(fēng)速在0.9~4.5m/s時(shí),標(biāo)靶球到地面三維激光掃描儀距離1 000~ 12 000 mm時(shí),平均配準(zhǔn)殘差為1.3 mm,個(gè)別殘差點(diǎn)位于4.5~5 mm之間。分析產(chǎn)生的原因?yàn)楣麍@自然環(huán)境中含有了灰塵或者飛蟲(chóng)的掃描點(diǎn),對(duì)配準(zhǔn)殘差產(chǎn)生了影響。平均擬合誤差為0.95 mm,最大擬合誤差小于3 mm。配準(zhǔn)殘差和擬合誤差均低于大場(chǎng)景測(cè)量配準(zhǔn)誤差要求(5 mm)[31]。標(biāo)靶球的擬合誤差變化說(shuō)明,在果園自然環(huán)境下掃描時(shí),蘋(píng)果樹(shù)冠層枝干和果實(shí)被視為具有剛體特性的目標(biāo)物,受外界環(huán)境變化較小,在配準(zhǔn)好的三維點(diǎn)云數(shù)據(jù)中,不同站點(diǎn)云重疊部分分布均勻。
蘋(píng)果樹(shù)冠層結(jié)構(gòu)中含有枝干、果實(shí)、葉片等植物器官,其中枝干和果實(shí)具有剛體特性,葉片因自身柔軟易發(fā)生變形,不具有剛體特性。為了實(shí)現(xiàn)掃描后三維點(diǎn)云數(shù)據(jù)的快速精確配準(zhǔn),在數(shù)據(jù)配準(zhǔn)中不區(qū)分掃描目標(biāo)物是否具備剛體特性。有風(fēng)果園環(huán)境中掃描時(shí),蘋(píng)果樹(shù)冠層葉片在掃描的瞬間會(huì)發(fā)生形態(tài)變化,獲取葉片的三維點(diǎn)云數(shù)據(jù)出現(xiàn)失真現(xiàn)象。葉片因自身材料特性使得地面三維激光掃描獲取的不同站三維點(diǎn)云數(shù)據(jù)不重疊的現(xiàn)象稱為分層現(xiàn)象。出現(xiàn)分層的葉片三維點(diǎn)云數(shù)據(jù)空間分布比實(shí)際葉片大,且不均勻,邊緣輪廓清晰程度不同。圖6和圖7列出了蘋(píng)果樹(shù)冠層的三維點(diǎn)云數(shù)據(jù)、部分枝干三維點(diǎn)云數(shù)據(jù)、葉片三維點(diǎn)云數(shù)據(jù),分析了不同風(fēng)速影響下,地面三維激光掃描儀獲取的蘋(píng)果樹(shù)冠層參數(shù)的不同質(zhì)量。
2.1 風(fēng)速與枝干、果實(shí)點(diǎn)云質(zhì)量關(guān)系分析
圖6是平均風(fēng)速4.5 m/s時(shí),部分枝干三維點(diǎn)云數(shù)據(jù)配準(zhǔn)后的圖像,棕色和綠色分別表示不同站掃描時(shí)獲取的三維點(diǎn)云,棕色和綠色交接處出現(xiàn)了棕色和綠色三維點(diǎn)云交叉現(xiàn)象,即為兩站配準(zhǔn)時(shí)重疊部分的三維點(diǎn)云數(shù)據(jù),圖像中交叉分布均勻,配準(zhǔn)后的點(diǎn)云數(shù)據(jù)樹(shù)干外形輪廓清晰。
其中圖6中為0.9、1.2、4.5 m/s 3種不同平均風(fēng)速環(huán)境中獲取的蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù),整體冠層數(shù)據(jù)比較清晰,冠層枝條、葉片、果實(shí)清晰可辨,點(diǎn)云密集??梢杂萌斯そ换シ椒ň_地提取樹(shù)干高度、冠層高度、胸徑、枝干夾角、果徑幾何參數(shù)信息。
a. 風(fēng)速Wind speed 0.9 m·s-1
b. 風(fēng)速Wind speed 1.2 m·s-1
c. 風(fēng)速Wind speed 4.5 m·s-1
注:從左到右:整體冠層、部分枝干、彩色信息、點(diǎn)云切片、參數(shù)提取
Note: From left to right: whole canopy, piece of branch, color data, point cloud slice, parameter extraction
圖6 蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)
Fig. 6 3D point cloud data of apple tree canopy
2.2 風(fēng)速與葉片點(diǎn)云質(zhì)量關(guān)系分析
圖7是在不同風(fēng)速下蘋(píng)果樹(shù)冠層部分葉片三維點(diǎn)云數(shù)據(jù)。表中可以看出,平均風(fēng)速為0、0.9、1.1、1.9、2.4 m/s時(shí),測(cè)量葉片側(cè)面三維點(diǎn)云數(shù)據(jù)分布寬度,作為地面三維激光掃描儀掃描的葉片側(cè)面厚度值,分別為0.9、5.5、9.8、14.5、35.8 mm。葉片側(cè)面厚度的不同反應(yīng)了地面三維激光掃描儀獲取的葉片三維點(diǎn)云的分層程度。
從圖7可以看出,在平均風(fēng)速為0的測(cè)量環(huán)境下,葉片側(cè)面的三維點(diǎn)云厚度為0.9 mm,三維點(diǎn)云數(shù)據(jù)分布均勻,生成三角網(wǎng)格大小均勻[32]。平均風(fēng)速在0.9~2.4 m/s之間變化時(shí),葉片側(cè)面的三維點(diǎn)云厚度變化區(qū)間為5.5~35.8 mm。平均風(fēng)速為0.9 m/s,配準(zhǔn)后的葉片三維點(diǎn)云分布比較均勻,表現(xiàn)在點(diǎn)云圖像中不同顏色的點(diǎn)云(不同站點(diǎn)云顏色不同)分布比較均勻,配準(zhǔn)的多站三維點(diǎn)云之間沒(méi)有出現(xiàn)分層現(xiàn)象。葉片側(cè)面厚度為5.5 mm,比實(shí)際葉片厚度值大,可以清晰分辨出葉片邊緣、尖端輪廓,原始點(diǎn)云生成的Delaunay三角網(wǎng)格比較均勻,沒(méi)有狹長(zhǎng)的三角形,易于提取葉片表型參數(shù)。平均風(fēng)速1.1 m/s時(shí),葉片開(kāi)始出現(xiàn)不同程度的分層現(xiàn)象,平均風(fēng)速1.9 m/s時(shí)的分層厚度為14.5 mm,葉片邊緣比較清晰,但是生成Delaunay三角網(wǎng)格時(shí),出現(xiàn)了狹長(zhǎng)三角形現(xiàn)象,不利于葉片表面信息提取。平均風(fēng)速2.4 m/s時(shí),三維點(diǎn)云數(shù)據(jù)從葉片側(cè)面看,葉片分層現(xiàn)象更為明顯,從葉片正面看,單站三維點(diǎn)云數(shù)據(jù)輪廓不清晰,多站三維點(diǎn)云數(shù)據(jù)配準(zhǔn)后沒(méi)有出現(xiàn)不同站點(diǎn)云重疊現(xiàn)象,生成的Delaunay三角網(wǎng)格沒(méi)有規(guī)律,不能提葉片邊緣輪廓等。
注:從左到右:葉片側(cè)面厚度、葉片正面、三角網(wǎng)絡(luò)。
在相同風(fēng)速下,地面三維激光掃描儀獲取的葉片三維點(diǎn)云數(shù)據(jù)的葉片側(cè)面厚度值是不同的,一方面可能是由測(cè)量誤差引起的,另一方面的主要可能是不同站掃描時(shí),瞬間風(fēng)速變化導(dǎo)致的。后者可以考慮采用多掃描儀協(xié)同掃描以減少風(fēng)速不均勻造成的葉片分層影響。
為了分析有風(fēng)的天氣對(duì)果樹(shù)冠層葉片三維點(diǎn)云數(shù)據(jù)質(zhì)量的影響程度,在不同風(fēng)速大小的天氣情況下掃描果樹(shù)冠層,提取三維點(diǎn)云數(shù)據(jù)中距離地面三維激光掃描儀不同距離的中等大小的葉片80枚,分析葉片三維點(diǎn)云數(shù)據(jù)不同風(fēng)速影響下,葉片的分層情況。采用曲線擬合方法,建立風(fēng)速與葉片側(cè)面厚度之間的曲線估計(jì)模型如圖8所示。
從圖8可以看出,平均風(fēng)速與葉片側(cè)面厚度的Quadratic曲線擬合模型、Cubic曲線擬合模型、Exponential曲線擬合模型的決定系數(shù)分別為0.976、0.986和0.983,<0.001,說(shuō)明3種曲線具有較好的擬合效果。
從Exponential曲線擬合模型可以看出,葉片側(cè)面厚度隨著風(fēng)速的增大而增大,Quadratic曲線擬合模型、Cubic曲線擬合模型分別在平均風(fēng)速1.22、1.20 m/s時(shí),葉片側(cè)面厚度出現(xiàn)最小值,在1.6 m/s時(shí),葉片厚度出現(xiàn)急速增大的趨勢(shì)。平均風(fēng)速在1.6 m/s以下時(shí),果園環(huán)境下獲得的三維點(diǎn)云數(shù)據(jù)可以獲取葉片邊緣輪廓和表型參數(shù)。
在果園環(huán)境下,采用地面三維激光掃描儀采集蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù),除了受到風(fēng)速的影響之外,還有溫度、光照條件、風(fēng)力、氣壓、空氣質(zhì)量等因素。另外,蘋(píng)果樹(shù)冠層包含的枝干、葉片、果實(shí)等材質(zhì)不同,使得實(shí)體本身反射特征不均勻,導(dǎo)致最終獲取的三維點(diǎn)云數(shù)據(jù)包含有錯(cuò)點(diǎn)和漏點(diǎn),在點(diǎn)云質(zhì)量上表現(xiàn)為枝干或者葉片等出現(xiàn)點(diǎn)云空洞,如圖6藍(lán)色圓所示。
2.3 測(cè)量精度分析
表2是2016年4月份新稍生長(zhǎng)監(jiān)測(cè)中測(cè)試的部分?jǐn)?shù)據(jù)(風(fēng)速1.4 m/s),蘋(píng)果樹(shù)冠層同一個(gè)參數(shù)取采用人工測(cè)量方法和三維點(diǎn)云數(shù)據(jù)測(cè)量方法均測(cè)量3次取平均值,對(duì)比2種蘋(píng)果樹(shù)冠層參數(shù)測(cè)量方法,以對(duì)比值,即人工測(cè)量值為標(biāo)準(zhǔn),測(cè)量精度高于人工測(cè)量,相對(duì)誤差小于4%。
表2 人工和三維點(diǎn)云測(cè)量參數(shù)的結(jié)果對(duì)比
注:風(fēng)速為1.4 m·s-1。
Note: The wind speed was 1.4 m·s-1.
1)提出了利用以地面三維激光掃描儀為采集設(shè)備的果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)精準(zhǔn)獲取方法。
2)以果園自然環(huán)境成長(zhǎng)的蘋(píng)果樹(shù)冠層為研究對(duì)象,提出了1種基于標(biāo)靶球的KD-trees-ICP算法,每站標(biāo)靶球單獨(dú)參與配準(zhǔn)。利用提出的算法,在后續(xù)試驗(yàn)中得到較好的三維點(diǎn)云配準(zhǔn)效果。果園環(huán)境下的蘋(píng)果樹(shù)冠層掃描試驗(yàn)表明,平均風(fēng)速在0.9~4.5 m/s時(shí),標(biāo)靶球距離地面三維激光掃描儀為1 000~12 000 mm時(shí),平均配準(zhǔn)殘差為1.3 mm,個(gè)別殘差點(diǎn)位于4.5~5 mm之間,平均擬合誤差為0.95 mm,最大擬合誤差小于3 mm,均小于大場(chǎng)景測(cè)量誤差要求的5 mm。
3)通過(guò)風(fēng)速對(duì)蘋(píng)果樹(shù)冠層三維點(diǎn)云影響分層試驗(yàn),結(jié)果表明,冠層中的枝干和果實(shí)在平均風(fēng)速4.5 m/s時(shí),掃描的三維點(diǎn)云數(shù)據(jù)沒(méi)有受到影響,蘋(píng)果樹(shù)冠層三維點(diǎn)云中的葉片出現(xiàn)了不同程度的分層現(xiàn)象,平均風(fēng)速越大分層現(xiàn)象越明顯,平均風(fēng)速小于1.6 m/s時(shí),葉片分層對(duì)葉片輪廓和表型參數(shù)提取影響不大。果園環(huán)境下,平均風(fēng)速小于1.6 m/s時(shí)可以從蘋(píng)果樹(shù)冠層三維點(diǎn)云數(shù)據(jù)中提取高精度冠層參數(shù)。同時(shí)與人工測(cè)量相比,利用地面激光三維掃描儀獲取12 000 mm以內(nèi)蘋(píng)果樹(shù)冠層參數(shù)結(jié)果,相對(duì)誤差小于4%。
[1] Miranda-Fuentes A, Llorens J, Gamarra-Diezma J L, et al. Towards an optimized method of olive tree crown volume measurement[J]. Sensors, 2015, 15(2): 3671-3687.
[2] Lichstein J W, Dushoff J, Ogle K, et al. Unlocking the forest inventory data: Relating individual tree performance to unmeasured environmental factors[J]. Ecol Appl, 2010, 20, 684-699.
[3] Vos J, Marcelis L F M, De Visser P H B, et al. Functional- structural plant modelling in crop production[C]. Springer Publishing Company, Incorporated, 2007: vii-viii(2).
[4] Piboule A, Collet C, Frochot H, et al. Reconstructing crown shape from stem diameter and tree position to supply light models. I. Algorithms and comparison of light simulations[J]. Annals of Forest Science, 2005, 62(7): 645-657.
[5] 董天陽(yáng),張萃,范菁,等. 植物共享光照資源的生長(zhǎng)模型改進(jìn)及可視化仿真[J]. 農(nóng)業(yè)工程學(xué)報(bào),2012,28(16):145-152. Dong Tianyang, Zhang Cui, Fan Jing, et al. Growth model improvement and visual simulation of plants sharing light resource[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2012, 28(16): 145-152. (in Chinese with English abstract)
[6] Chen Y, Zhu H, Ozkan H E, et al. Spray drift and off-target loss reductions with a precision air-assisted sprayer[J]. Transactions of the ASABE, 2013, 56(6): 1273-1281.
[7] 蘇渤海,范崇輝,李國(guó)棟,等. 紅富士蘋(píng)果改形過(guò)程中不同樹(shù)形光照分布及其對(duì)產(chǎn)量品質(zhì)的影響[J]. 西北農(nóng)林科技大學(xué)學(xué)報(bào): 自然科學(xué)版,2008,36(1):158-162. Su Bohai, Fan Conghui, Li Guodong, et al. Effects of modifying between light distribution, yield and quality of different shapes on "Red Fuji" apple[J]. Journal of Northwest A&F University: Natural science, 2008, 36(1): 158-162. (in Chinese with English abstract)
[8] Tong J, Zhang Q, Karkee M, et al. Understanding the dynamics of hand picking patterns of fresh market apples[C]. American Society of Agricultural and Biological Engineers, 2014 Montreal, Quebec Canada, 2014.
[9] 買(mǎi)買(mǎi)提艾沙江,木米尼阿布都外力,王晶晶,等. 富士蘋(píng)果短截,拉枝對(duì)當(dāng)年生新梢葉片光合特性的影響[J]. 中國(guó)農(nóng)業(yè)大學(xué)學(xué)報(bào),2013 (6):126-131. Aishajiang M, Abuduwaili M, Wang Jingjing, et al. Effects of cutting back and branch-bending on photosynthetic characteristics of new shoots[J]. Journal of China Agricultural University, 2013, 6: 018. (in Chinese with English abstract)
[10] Khosravipour A, Skidmore A K, Wang T, et al. Effect of slope on treetop detection using a LiDAR canopy height model[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 104: 44-52.
[11] Bargoti S, Underwood J P, Nieto J I, et al. A pipeline for trunk detection in trelliss tructured apple orchards[J]. Journal of Field Robotics, 2015, 32(8): 1075-1094.
[12] Str?mbu V F, Str?mbu B M. A graph-based segmentation algorithm for tree crown extraction using airborne LiDAR data[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 104: 30-43.
[13] Miranda-Fuentes A, Llorens J, Gamarra-Diezma J L, et al. Towards an optimized method of olive tree crown volume measurement[J]. Sensors, 2015, 15(2): 3671-3687.
[14] 劉芳,馮仲科,楊立巖,等. 基于三維激光點(diǎn)云數(shù)據(jù)的樹(shù)冠體積估算研究[J]. 農(nóng)業(yè)機(jī)械學(xué)報(bào),2016,47(3):328-334. Liu Fang, Feng Zhongke, Yang Liyan, et al. Estimation of tree crown volume based on 3D laser point clouds data[J]. Transactions of the Chinese Society for Agricultural Machinery, 2016, 47(3): 328-334. (in Chinese with English abstract)
[15] Leeuwen M V, Nicholas C C, Thomas H, Michael A W, et al. Automated reconstruction of tree and canopy structure for modeling the internal canopy radiation regime[J]. Remote Sensing of Environment, 2013, 136: 286-300.
[16] Schneider F D, Leiterer R, Morsdorf F, et al. Simulating imaging spectrometer data: 3D forest modeling based on LiDAR and in situ data[J]. Remote Sensing of Environment, 2014, 152: 235-250.
[17] 高士增,張懷清,劉閩,等. 基于點(diǎn)云的樹(shù)木枝干形態(tài)參數(shù)提取技術(shù)[J].東北林業(yè)大學(xué)學(xué)報(bào),2014(4):109-114. Gao Shizeng, Zhang Huaiqing, Liu Min, et al. Morphological parameters extraction of tree branches based on point cloud[J]. Journal of Northeast Forestry University, 2014(4): 109-114. (in Chinese with English abstract)
[18] 蘇紅波,郭新宇,陸聲鏈,等. 蘋(píng)果花序幾何造型及可視化研究[J]. 中國(guó)農(nóng)學(xué)通報(bào),2009(2): 272-276. Su Hongbo, Guo Xinyu, Lu Shenglian, et al. Geometry modelling and visualization of the apple inflorescence[J]. Chinese Agricultural Science Bulletin, 2009(2): 272-276. (in Chinese with English abstract)
[19] 方慧,杜朋朋,胡令潮,等. 基于可視化類庫(kù)的植株三維形態(tài)配準(zhǔn)方法及點(diǎn)云可視化[J]. 農(nóng)業(yè)工程學(xué)報(bào),2013,29(22):180-188. Fang Hui, Du Pengpeng, Hu Lingchao, et al. VTK-based plant 3D morphological visualization and registration[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2013, 29(22): 180-188. (in Chinese with English abstract)
[20] 郭浩,朱德海,戈振揚(yáng),等. 植物立體可視化模擬系統(tǒng)原型StereoPlants的構(gòu)建[J]. 農(nóng)業(yè)工程學(xué)報(bào),2012,28(15):113-117. Guo Hao, Zhu Dehai, Ge Zhenyang, et al. Construction of plant stereo visualization simulation system prototype: Stereo plants[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2012, 28(15): 113-117. (in Chinese with English abstract)
[21] 趙春江,陸生鏈,郭新宇,等. 數(shù)字植物研究進(jìn)展:植物形態(tài)結(jié)構(gòu)三維數(shù)字化[J]. 中國(guó)農(nóng)業(yè)科學(xué),2015,48(17):3415-3428. Zhao Chunjiang, Lu Shenglian, Guo Xinyu, et al. Advances in research of digital plant: 3D digitization of plant morphological structure[J]. Scientia Agricultura Sinica, 2015, 48(17): 3415-3428. (in Chinese with English abstract)
[22] 程效軍,施貴剛,王峰,等.點(diǎn)云配準(zhǔn)誤差傳播規(guī)律的研究[J]. 同濟(jì)大學(xué)學(xué)報(bào):自然科學(xué)版,2009,37(12):1668-1672. Cheng Xiaojun, Shi Guigang, Wang Feng, et al. Research on point cloud registration error propagation[J].Journal of Tongji University: Natural Science, 2009, 37(12): 1668-1672. (in Chinese with English abstract)
[23] Besl P J, McKay N D. Method for registration of 3-D shapes[C]// International Society for Optics and Photonics : Robotics-DL tentative, 1992: 586-606.
[24] Chen Y, Medioni G. Object modelling by registration of multiple range images[J]. Image and Vision Computing, 1992, 10(3): 145-155.
[25] Vo A V, Truong-Hong L, Laefer D F, et al. Octree-based region growing for point cloud segmentation[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 104: 88-100.
[26] Phillips J M, Liu R, Tomasi C. Outlier robust ICP for minimizing fractional RMSD[C]// IEEE Computer Society: International Conference on 3-D Digital Imaging and Modeling, 2007:427-434..
[27] 孫軍華,謝萍,劉震,等. 基于分層塊狀全局搜索的三維點(diǎn)云自動(dòng)配準(zhǔn)[J]. 光學(xué)精密工程,2013(1):174-180. Sun Junhua, Xie Ping, Liu Zhen, et al. Automatic 3 D point cloud registration based on hierarchical block global search[J]. Optic and Precision Engineer, 2013, 21(1): 174-180. (in Chinese with English abstract)
[28] Greenspan M, Yurick M. Approximate kd tree search for efficient ICP[C]//International Conference on 3-D Digital Imaging and Modeling, 2003. 3dim 2003. 2003: 442-448.
[29] Rusu R B, Marton Z C, Blodow N, et al. Towards 3D Point cloud based object maps for household environments[J]. Robotics & Autonomous Systems, 2008, 56(11): 927-941.
[30] 康志忠,王薇薇,李珍. 多源數(shù)據(jù)融合的三維點(diǎn)云特征面分割和擬合一體化方法[J]. 武漢大學(xué)學(xué)報(bào):信息科學(xué)版,2013(11):1317-1321,1382. Kang Zhizhong, Wang Weiwei, Li Zhen. An Integrative Method of 3D Point Clouds Feature Segmentation and Fitting Fusing Multiple Data Sources[J]. Geomatics and Information Science of Wuhan University, 2013(11): 014. (in Chinese with English abstract)
[31] 尹若捷,楊沖,付正文.多激光掃描儀的大場(chǎng)景聯(lián)合配準(zhǔn)及 精度評(píng)估[J]. 測(cè)繪地理信息,2015(3):13-17. Yin Ruojie, Yang Chong, Fu Zhengwen. Overall control and registration precision analysis of large-scale terrain with multiple laser scanners[J]. Journal of Geomatics, 2015(3): 13-17. (in Chinese with English abstract)
[32] Hou T, Zheng B, Xu Z, et al. Simplification of leaf surfaces from scanned data: Effects of two algorithms on leaf morphology[J]. Computers and Electronics in Agriculture, 2016, 121: 393-403.
Apple tree canopy geometric parameters acquirement based on 3D point clouds
Guo Cailing1,2, Zong Ze1, Zhang Xue1, Liu Gang1※
(1.100083,;2.063000,)
Accurate structural parameters and crown characterization of large isolated apple trees were vital for adjusting spray doses, trimming, autonomous harvesting. According to previous research, canopy measurement methods to characteristic the whole tree structure could be classified in two groups: Manual measurements and electronic procedures to estimate tree dimensions. These methods were time consuming and required specialist knowledge, so a simpler crown characterization measurement method was required. Terrestrial laser scanning (TLS) could provide accurate canopy information through non-destructive methods, which filled the gap between tree scale manual measurements and large scale LiDAR measurements. Laser scanning delivers a dense cloud of points, and this raw point data were filtered for deriving a digital terrain model and subsequent fitting of a parametric stem model. In this study, Trimble TX8 had been used to getting point clouds of the apple tree canopy with trees height 3.2-5.1 m and 7 years old, in the orchard environment. A method and registration algorithm for reconstructing the three-dimensional (3D) apple tree canopy based on terrestrial laser scanner point cloud data from apple trees was presented. After an initial alignment had been obtained from this last set of correspondences, the object ball point clouds were extracted, and the noise was deleted by hands. In order to improve convergence speed, KD-tree improved ICP(iterated closest points), and combined with object ball, to estimate the optimal transform. The object residual errors and fitting errors at different distances between object and scanner were analyzed. Results showed that, the average residual errors was 1.3 mm, and the average fitting errors was 0.95 mm at the distance from 1 000 to 13 000 mm. All the errors were less than the traditional registration accuracy 5 mm. In addition, wind as an importance factor always influenced point clouds quality. In order to find the influence between them, several pieces of branches, apples and 80 pieces of leaves had been extracted in the wind speed from 0.9 to 4.5 m/s. And the branches and apple structures, the leaf characteristics were studied under different wind speed. Results showed that, the branches and apple outline clearly, both the single tree and group trees, the geometric parameters, such as apple diameter, stem diameter, trunk detection, canopy height, canopy diameter, planting distance, line spacing, could been extracted easily even if the average wind speed was 4.5m/s in the scanning instant. Great changes had taken place in the leaves edge and thickness, when the wind speed changed from 0.9 to 2.4 m/s. The thickness of the leaf profile had changed from 2.2 to about 35.8 mm, and the original point clouds Delaunay triangular mesh also became irregular. And long and narrow triangle appeared at the moment of the average wind speed 1.9 m/s. The three leaf thickness fitting curves, as quadratic curve, cubic curve and exponential curve, were in good agreements for the whole range of studied volumes (2=0.976,2=0.986 and2= 0.983,< 0.001). The fitting curve showed that, apple canopy 3D point cloud data could be obtained with good quality in orchard environment. Comparing with the traditional manual measurement, the relative errors of the canopy parameter measurement values obtained from 3D point clouds data were less than 4%.
measurement errors; lasers; accuracy; apple canopy; point cloud
10.11975/j.issn.1002-6819.2017.03.024
F762.5;P225
A
1002-6819(2017)-03-0175-07
2016-06-16
2016-12-29
國(guó)家自然科學(xué)基金資助項(xiàng)目(31371532)
郭彩玲,女,河北石家莊人,副教授,博士生,主要從事自動(dòng)化與信息化技術(shù)研究;北京中國(guó)農(nóng)業(yè)大學(xué),100083;唐山唐山學(xué)院,063000。Email:gcl@cau.edu.cn
劉 剛,男,河北保定人,教授,博士生導(dǎo)師,主要從事電子信息技術(shù)在農(nóng)業(yè)中的應(yīng)用研究;北京中國(guó)農(nóng)業(yè)大學(xué),100083。 Email:pac@cau.edu.cn
郭彩玲,宗 澤,張 雪,劉 剛.基于三維點(diǎn)云數(shù)據(jù)的蘋(píng)果樹(shù)冠層幾何參數(shù)獲取[J]. 農(nóng)業(yè)工程學(xué)報(bào),2017,33(3):175-181. doi:10.11975/j.issn.1002-6819.2017.03.024 http://www.tcsae.org
Guo Cailing, Zong Ze, Zhang Xue, Liu Gang.Apple tree canopy geometric parameters acquirement based on 3D point clouds[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2017, 33(3): 175-181. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2017.03.024 http://www.tcsae.org