sensors

Article Verification of Geometric Model-Based Phenotyping Methods for Studies of Xerophytic

Paweł Drapikowski 1,*, Ewa Kazimierczak-Grygiel 2, Dominik Korecki 1 and Justyna Wiland-Szyma ´nska 2,3

1 Institute of Control and Information Engineering, Faculty of Electrical Engineering, Poznan University of Technology, Piotrowo 3a, 60-965 Poznan, Poland; [email protected] 2 Botanical Garden, Adam Mickiewicz University, D ˛abrowskiego 165, 60-594 Poznan, Poland; [email protected] (E.K.-J.); [email protected] (J.W.-S.) 3 Department of Plant , Faculty of Biology, Adam Mickiewicz University, Umultowska 89, 61-614 Poznan, Poland * Correspondence: [email protected]; Tel.: +48-616652874

Academic Editor: Simon X. Yang Received: 16 February 2016; Accepted: 31 May 2016; Published: 27 June 2016

Abstract: This paper presents the results of verification of certain non-contact measurement methods of plant scanning to estimate morphological parameters such as length, width, area, volume of and/or stems on the basis of computer models. The best results in reproducing the shape of scanned objects up to 50 cm in height were obtained with the structured-light DAVID Laserscanner. The optimal triangle mesh resolution for scanned surfaces was determined with the measurement error taken into account. The research suggests that measuring morphological parameters from computer models can supplement or even replace phenotyping with classic methods. Calculating precise values of area and volume makes determination of the S/V (surface/volume) ratio for cacti and other succulents possible, whereas for classic methods the result is an approximation only. In addition, the possibility of scanning and measuring plant which differ in morphology was investigated.

Keywords: plant digitalization; morphological parameters; DAVID laser scanning system

1. Introduction Plant phenotyping is the comprehensive assessment of complex plant traits including their architecture. Classical methods are used to study variability, adaptation to environmental conditions and pace of growth during the vegetation period. So far, plant phenotyping has been mainly limited to crop plants or the model plant Arabidopsis thaliana (L.) Heynh [1]. In many cases, classical biometric measurement methods are sufficient. However, there are plants whose complex spatial structure and susceptibility to damage make employing conventional measurement techniques impossible or exceedingly difficult. Therefore, there is a need for new platforms and solutions that enable phenotypic evaluation of such species. In such cases, non-contact measurement techniques need to be used: preparation of a three-dimensional (3D) computer model (digitalization) and capturing plant traits based on it [1,2]. One group of plants that has not been used as an object for 3D phenotyping is xerophytes. Representatives of this remarkable group are interesting for several reasons. These plants show various morphological, anatomical and physiological adaptations to living in dry conditions and in sustaining an intensive herbivory [3–7]. They constitute the main vegetation in some desert and semi-desert

Sensors 2016, 16, 924; doi:10.3390/s16070924 www.mdpi.com/journal/sensors Sensors 2016, 16, 924 2 of 14 environments and are therefore a subject of intensive ecological study [8]. Their medicinal properties Sensors 2016, 16, 924 2 of 14 are recognized by both traditional and contemporary medicine as well as veterinary science [9–12]. Xerophytesmedicinal properties are also are desirable recognized ornamental by both plantstraditional worldwide. and contemporary The demand medicine for these as well plants as is so highveterinary that their science existence [9–12]. in the wild is threatened [13]. Therefore, many xerophytes, including the whole cactiXerophytes family and are also genera desirableAloe L.ornamental and plants worldwide.Lindl [14 The,15 ],demand are protected for these by plants international is so law underhigh that the their Washington existence in Convention the wild is threaten (CITES)ed as [13]. well Therefore, as by national many xerophytes, laws. Therefore, including customsthe officerswhole must cacti be family able to and recognize genera them L.properly and Pachypodium to prevent Lindl illegal [14,15], traffic. are protected The diversity by international of these plants law under the Washington Convention (CITES) as well as by national laws. Therefore, customs is, however, so great that only specialists are able to recognize them and classify them accordingly. officers must be able to recognize them properly to prevent illegal traffic. The diversity of these Comprehensive virtual guides that would enable rapid and simple identification of these plants do plants is, however, so great that only specialists are able to recognize them and classify them not exist.accordingly. Comprehensive virtual guides that would enable rapid and simple identification of Climatethese plants change do not as exist. well population growth have created an urgent need for sustainable high-yield crop production.Climate change Phenotypic as well studies population of plants growth adapted have to created difficult an environmental urgent need for conditions sustainable can be helpfulhigh-yield in finding crop new production. ways for cropPhenotypic improvement studies [of16 ].plants adapted to difficult environmental Variousconditions types can be of helpful scanners in finding make non-contactnew ways for plant crop improvement digitalization [16]. possible. Most often they work by castingVarious structured types light of scanners in various make patterns non-contact onto plant the objectdigitalization and recording possible. theMost images often they (Figure work1)[ 17]. On theby basiscasting of structured disruptions light in in the various pattern patterns caused onto by the the object object’s and surface,recording a the three-dimensional images (Figure 1) point [17]. On the basis of disruptions in the pattern caused by the object’s surface, a three-dimensional cloud is determined, which is further triangulated into the form of a triangle mesh. The triangle mesh point cloud is determined, which is further triangulated into the form of a triangle mesh. The computer model is then processed to determine morphological parameters of interest. triangle mesh computer model is then processed to determine morphological parameters of interest.

Figure 1. Scanning a in the pot with the structured light-based DAVID FigureLaserscanner 1. Scanning [17]. a succulent plant in the pot with the structured light-based DAVID Laserscanner [17]. Non-contact measurements of selected characteristics are possible using the Ubiquitous Sensor Non-contactNetwork proposed measurements by Suk et al. of [18]. selected An example characteristics of image-based are possible plant usingphenotyping the Ubiquitous on a much Sensor Networklarger proposed scale wasby presented Suk et al.by [18Kircherer]. An example et al. [19], of and image-based numerous additional plant phenotyping examples can on abe much found larger [1,2,20–23]. scale was presented by Kircherer et al. [19], and numerous additional examples can be found [1,2,20– Scanning larger objects is feasible with the use of, for instance, a Kinect scanner [24]. Recently, 23]. the photogrammetric method for creating computer models has been gaining popularity. It uses a Scanningpreviously largercaptured objects set of is images. feasible There with are the no use limits of, forto the instance, size of objects: a Kinect the scanner only restriction [24]. Recently, is the the photogrammetricability to take a method photograph. for creatingProfessional computer photographic models equipment has been is not gaining required, popularity. only a digital It uses a previouslycamera is captured necessary set [25]. of These images. scanning There methods are no limits require to relatively the size oflow-cost objects: devices the only which restriction offer is the abilityacceptable to take metrological a photograph. characteristics. Professional Extensive photographic metrological equipment and usability is not descriptions required, onlyof those a digital cameradevices is necessary for plant measurement [25]. These scanninghave been published methods [2 require6,27]. Better relatively results low-cost in scanning devices plants whichof very offer acceptablecomplex metrological shapes havecharacteristics. been achieved with Extensive hand-h metrologicaleld scanners, ande.g., usabilityArtec S [28] descriptions or HandyScan of those devicesEXASCAN for plant [29]. measurement Also, triangulation have been scanners published (typically [26 ,27more]. Better accurate), results mounted in scanning on special plants of measuring arms [30,31], can be used for this purpose, e.g., Artec S and Perceptron ScanWorks V5 very complex shapes have been achieved with hand-held scanners, e.g., Artec S [28] or HandyScan with the Romer Infinite 2.0 arm. However, their prices are far higher than those of hand-held EXASCANdevices. [ 29]. Also, triangulation scanners (typically more accurate), mounted on special measuring arms [30,31], can be used for this purpose, e.g., Artec S and Perceptron ScanWorks V5 with the Romer Infinite 2.0 arm. However, their prices are far higher than those of hand-held devices.

Sensors 2016, 16, 924 3 of 14

One frequently measured parameter in plant biology is S/V (surface to volume ratio). It is easy to determine it for a flat-leaved plant when the surface and thickness of a are measured. It is also possible to measure in situ with a portable scanner, e.g., Portable Living Leaf Area Meter [32]. For geometrically more complex plants, determination of S/V requires calculation of the surface and volume. The present work verifies certain non-contact measurement methods of certain plant morphological parameters (length, width, area and volume of leaf) that are otherwise very difficult or impossible to determine with traditional techniques. The paper also describes difficulties in plant scanning due to complicated surfaces, the presence of thorns or prickles that diffuse light, as well as hard-to-reach fragments that make it impossible to recreate the surface.

2. Materials and Methods

2.1. Materials The research was conducted on specimens belonging to nine species of xerophytic plants grown in the Botanical Garden of the Adam Mickiewicz University in Poznan, Poland. The species are representatives of various ecological groups of xerophytes. They are all included in the CITES Appendices (Table1). The families and names of species were applied according to the TROPICOS database [33].

Table 1. List of plant specimens used in the study.

Family Species AMU BG Accession Number CITES Appen. Ecological Form Aloaceae A. Berger I_I004_002_0000_6010_4570 II leaf succulent Pachypodium lamerei Drake I_I004_002_0000_6997_3368 II stem succulent Astrophytum capricorne (A. Dietr.) Cactaceae I_I002_003_0000_6996_4580 II stem succulent Britton et Rose Cactaceae Astrophytum myriostigma Lem. I_I002_003_0000_6996_4581 II stem succulent Echinopsis leucantha (Gillies ex Cactaceae I_I002_003_0000_6996_4571 II stem succulent Salm-Dyck) Walp. Cactaceae Leuchtenbergia principis Hook. I_I002_003_0000_6997_4590 II stem succulent Cactaceae Mammillaria magnimamma Haw. I_I002_003_0000_6999_4573 II stem succulent Cactaceae Pilosocereus pachycladus F. Ritter I_I002_003_0000_6998_4572 II stem succulent Welwitschiaceae Welwitschia mirabilis Hook.f. I_I004_001_0000_6998_4560 I true xerophyte

2.2. Methods

2.2.1. Distance Measurement between Two Points A three-dimensional triangular mesh forms a computer model. There are two ways to measure the distance between two given points in such a model. The first is to run the shortest route between the given points along the edges of triangles forming the mesh. The mesh is transformed to a graph which has the distances as the weights. Dijkstra’s algorithm was used to find the shortest route in the graph (Figure2B) [ 34]. Dijkstra’s algorithm finds the shortest paths from the source node to all other nodes in the graph, producing a shortest-path tree. The algorithm stops when the destination node is determined. The second method is projection of a straight line which joins two points onto the surface of the object and in finding the shortest path outside the edges of the triangles (Figure2A). This latter method is more accurate, as it shows indeed the shortest route, but it is more difficult to implement and does not always yield the correct result. This technique is used by specialized 3D RapidForm 2006 software to process the scanned data [35]. Even so, when the surface is a complex geometric shape, the distance is either not calculated at all or is calculated incorrectly (blue line in Figure2C). An object Sensors 2016, 16, 924 4 of 14 depicted in Figure2C is only an example of an improper shortest-path estimation by commercially availableSensors software 2016, 16, 924 [35 ]. Finding the shortest path with Dijkstra’s method for the same object4 isof correct14 and shown in Figure2D. Red lines in Figure2A,C depict the direct connection between points but the Sensorsobject 2016is correct, 16, 924 and shown in Figure 2D. Red lines in Figure 2A,C depict the direct connection4 of 14 blue onesbetween depict points the but real the connection blue ones depict based the on real mesh. connection based on mesh. object is correct and shown in Figure 2D. Red lines in Figure 2A,C depict the direct connection between points but the blue ones depict the real connection based on mesh.

Figure 2. Finding the shortest path between points (part of an Aloe marlothii leaf): (A) A line Figure 2. Finding the shortest path between points (part of an Aloe marlothii leaf): (A) A line projection projection onto the surface; (B) Triangle mesh-based; (C) An example of the incorrect shortest path; onto the surface; (B) Triangle mesh-based; (C) An example of the incorrect shortest path; (D) The Figure(D) The 2. correct Finding triangle the shortestmesh-based path path. between points (part of an Aloe marlothii leaf): (A) A line correctprojection triangle onto mesh-based the surface; path. (B) Triangle mesh-based; (C) An example of the incorrect shortest path; 2.2.2.( DLeaf’s) The correctLength triangle Estimation mesh-based as a Center path. Line 2.2.2. Leaf’s Length Estimation as a Center Line 2.2.2.Welwitschia Leaf’s Length mirabilis Estimation is among as a Centerthe species Line that are difficult to measure with classic methods Welwitschiabecause of the mirabilis shape ofis the among leaves the and species susceptibility that are to difficult damage to (Figure measure 3A). with classic methods because Welwitschia mirabilis is among the species that are difficult to measure with classic methods of thebecause shape of the the shape leaves of andthe leaves susceptibility and susceptibility to damage to damage (Figure 3(FigureA). 3A).

Figure 3. Welwitschia mirabilis: (A) Natural picture; (B) Computer model. Figure 3. Welwitschia mirabilis: (A) Natural picture; (B) Computer model. The lengthFigure of the 3. leafWelwitschia is usually mirabilis measured:(A) Naturalalong the picture; line going (B) Computer through its model. middle. When the leaf is twisted, as is the case with Welwitschia mirabilis (Figure 3), Dijkstra’s algorithm does not give goodThe results length because of the the leaf shortest is usually path measured does not liealong in the the middle line going of the through leaf. To its determine middle. theWhen length the Theleafof the is length leaf,twisted, multiple of theas is leaf stepsthe iscase are usually with needed: Welwitschia measured mirabilis along the (Figure line going 3), Dijkstra’s through algorithm its middle. does When not give the leaf is twisted,good results as is the because case withthe shortestWelwitschia path does mirabilis not lie(Figure in the middle3), Dijkstra’s of the leaf. algorithm To determine does notthe length give good  finding the outer edges of the leaf, resultsof becausethe leaf, multiple the shortest steps pathare needed: does not lie in the middle of the leaf. To determine the length of the  mapping the distances from all vertices to the edges in the model using the Fast Marching leaf, multiple steps are needed:  findingMethod the (FMM) outer [36] edges (Figure of the 4A), leaf,

 mapping the distancesdistances betweenfrom all all vertices vertices to in the the edges triangular in the mesh model from using the startingthe Fast point Marching using ‚ finding the outer edges of the leaf, Methodthe FMM (FMM) method [36] and (Figure division 4A), into a given number of areas (Figure 4B,C),

‚ mapping mappingdetermining the distancesthe thedistances vertices from between that all vertices are all furthest vertices to the from in edges the the triangular inobject’s the model edgesmeshusing fromin every the the startingarea Fast except Marching point the using one Method (FMM)thecontaining [FMM36] (Figure method the starting4A), and division point and into the a endpointgiven number (Figure of areas4D), (Figure 4B,C),

‚ mapping determining the distances the verticesshortest between thatroutes are all between furthest vertices verticesfrom in the the triangularset object’s in the edgesprevious mesh in fromevery step,the areataking starting except into accountthe point one using containing the starting point and the endpoint (Figure 4D), the FMMthe starting method point and and division the endpoint, into a giventhen summing number them of areas (Figure (Figure 4E). 4B,C),  determining the shortest routes between vertices set in the previous step, taking into account ‚ determiningthe starting the point vertices and the that endpoint, are furthest then summing from the them object’s (Figure edges 4E). in every area except the one containing the starting point and the endpoint (Figure4D),

Sensors 2016, 16, 924 5 of 14

‚ determining the shortest routes between vertices set in the previous step, taking into account the starting point and the endpoint, then summing them (Figure4E). Sensors 2016, 16, 924 5 of 14

Figure 4. Steps to determine the length of a leaf: (A) Map of distances from the edge of the leaf; Figure 4. Steps to determine the length of a leaf: (A) Map of distances from the edge of the leaf; (B) Map (B) Map of distances from the starting point to the endpoint; (C) Area division; (D) Vertices furthest of distances from the starting point to the endpoint; (C) Area division; (D) Vertices furthest away from away from the edge; (E) The shortest path; (F) Comparison of the path through the middle of the leaf the edge; (E) The shortest path; (F) Comparison of the path through the middle of the leaf (red line) (red line) and the shortest path in the geometric sense (black line) [37]. and the shortest path in the geometric sense (black line) [37]. Figure 4F shows the paths obtained with both methods between the starting point and the endingFigure point.4F shows Red thepresents paths the obtained line put with through both methods the middle between of the the leaf starting while point black and presents the ending the point.shortest Red geometric presents path the line between put through those points the middle determined of the with leaf Dijkstra’s while black method. presents It works the shortest well geometricwhen the path object between is not geometrically those points complex, determined for instance with Dijkstra’s with Aloe method. marlothii Itleaves. works However, well when with the objectobjects is notsuch geometrically as leaves of complex,Welwitschia for mirabilis, instance it with becomesAloe marlothiinecessaryleaves. to use However,the multi-step with procedure objects such asdetailed leaves of above.Welwitschia mirabilis, it becomes necessary to use the multi-step procedure detailed above.

2.2.3.2.2.3. Errors Errors of of Measurement Measurement Based Based onon ComputerComputer Models TheThe use use of computerof computer models models requires requires verification verificati ofon precision of precision in their in creationtheir creation and in and the accuracyin the ofaccuracy measurements. of measurements. It is assumed It is that assumed the precision that the ofprecision models of is models based onis based manufacturer on manufacturer data which data for thewhich DAVID for 3Dthe SLA-1 DAVID scanner 3D SLA-1 is 0.1% scanner of the is object 0.1% size.of the Such object precision size. Such is sufficient precision for is measurementssufficient for ofmeasurements plants’ characteristics. of plants’ characteristics. Measurement Measurement accuracy based accuracy on triangle based mesh on triangle models mesh also models depends also on meshdepends resolution. on mesh Error resolution. of measurement Error of measurement of different characteristicsof different characteristics depending ondepending mesh resolution on mesh is resolution is calculated in relation to the densest mesh model taken as the reference (Model7, calculated in relation to the densest mesh model taken as the reference (Model7, Equation (1)). Equation (1)). ModelModel7 ´ Model n ErrorError“7 n ¨100%100% wherewhere n n P x 1,61, 6y (1) Model7 (1) Model7

DetailedDetailed analysis analysis and and results results are are presentedpresented inin SectionSection 3.13.1..

3.3. Results Results 3.1. Errors in Measurement of Distances, Areas and Volumes on the Basis of Computer Models 3.1. Errors in Measurement of Distances, Areas and Volumes on the Basis of Computer Models ErrorsErrors in in determining determining thethe shortestshortest distancedistance between the the points, points, the the area area or or the the volume volume in inthe the triangulartriangular mesh mesh model model depend depend onon thethe spatialspatial resolutionresolution of of the the mesh. mesh. Spatial Spatial resolution resolution is isset set in inthe the finalfinal stage stage of of model model creation creation in in the thescanning scanning software.software. In order order to to analyze analyze how how the the mesh mesh resolution resolution influencesinfluences measurement measurement accuracy,accuracy, thethe samesame modelmodel of the Aloe marlothii marlothii leafleaf was was processed processed with with differentdifferent resolutions resolutions understood understood as as the the mean mean distance distance between between the points.the points. Table Table2 presents 2 presents the results the ofresults measuring of measuring the distance, the area,distance, and volume.area, and The volume. number The of number points and of points triangles and indicates triangles that indicates Model 3 isthat approximately Model 3 is approximately 60 times “smaller” 60 times (in the“smaller” mean of(in the the number mean of of the points number and of triangles) points and than triangles) Model 7 butthan its errorModel is only7 but1% its witherror respect is only to1% Model with 7.respect Figure to5 depictsModel 7. the Figure relationship 5 depicts of measurementthe relationship errors of measurement errors to the reference model. A graphic representation is shown in Figure 6, where the red line depicts the distance between points according to Dijkstra’s algorithm. Models 5, 6, and 7 are not shown because the mesh density is such that the triangle mesh is not visible.

Sensors 2016, 16, 924 6 of 14

Sensors 2016, 16, 924 6 of 14 toSensors the reference 2016, 16, 924 model. A graphic representation is shown in Figure6, where the red line depicts6 of 14 the distance between points according to Dijkstra’s algorithm. Models 5, 6, and 7 are not shown because Table 2. Results of calculating the area, volume, and the distances between points depending on the the mesh density is such that the triangle mesh is not visible. resolutionTable 2. Results of the triangularof calculating mesh the used area, in volume, various and models. the distance s between points depending on the resolution of the triangular mesh used in various models. ModelTable Number 2. Results of calculating the area, volume, 1 and the distances2 between 3 points 4 depending 5 on 6 the 7 Model Number 1 2 3 4 5 6 7 Meanresolution distance ofbetween the triangular vertices mesh [mm] used in various 3.97 models. 2.03 1.09 0.56 0.28 0.20 0.14 Mean distance between vertices [mm] 3.97 2.03 1.09 0.56 0.28 0.20 0.14 Number of points 859 3442 12,013 48,159 191,352 370,824 713,645 Number of points 859 3442 12,013 48,159 191,352 370,824 713,645 NumberModel of Number triangles 1664 1 26852 24,022 3 96,314 4 382,700 5 741,632 6 1,427,286 7 Number of triangles 1664 6852 24,022 96,314 382,700 741,632 1,427,286 AreaMean [mm distance2] between vertices [mm]11,314 3.97 2.03 11,919 1.09 12,100 0.56 12,175 0.28 12,206 0.20 12,209 0.14 12,216 AreaNumber [mm of2] points11,314 859 3442 11,919 12,013 12,100 48,159 12,175191,352 12,206 370,824 12,209713,645 12,216 Error % 7.38 2.43 0.95 0.34 0.08 0.06 0.00 ErrorNumber % of triangles 1664 7.38 6852 2.43 24,022 0.95 96,314 0.34382,700 0.08 741,632 0.061,427,286 0.00 Area [mm2]3 11,314 11,919 12,100 12,175 12,206 12,209 12,216 VolumeVolume [mm [mm]3 ] 52,463 55,364 55,364 55,993 55,993 56 56,167,167 56,206 56,206 56,172 56,172 56,174 56,174 ErrorError % % 7.38 6.61 2.43 1.44 0.95 0.32 0.34 0.01 0.08 0.06 0.06 0.00 0.00 0.00 ErrorVolume % [mm3] 52,463 6.61 55,364 1.44 55,993 0.32 56,167 0.01 56,206 0.06 56,172 0.00 56,174 0.00 Distance–triangularDistance–triangularError % mesh mesh [mm] [mm] 6.61 179.7 1.44 184.5 184.5 0.32 181.7 181.7 0.01 183.8 183.8 0.06 183.8 183.8 0.00 183.1 183.1 0.00183.5 183.5 ErrorErrorDistance–triangular % % mesh [mm] 179.7 2.10 184.5 0.53 0.53 181.7 0.99 0.99 183.8 0.12 0.12 183.8 0.16 0.16 183.1 0.23 0.23 183.5 0.00 0.00 Distance–segmentError % projection [mm] 2.10 169.3 0.53 175.7 0.99 176.2 0.12 176.0 0.16 176.4 0.23 177.0 0.00 177.2 Distance–segmentDistance–segment projectionprojection [mm] [mm] 169.3 169.3 175.7 175.7 176.2 176.2 176.0 176.0 176.4 176.4 177.0 177.0 177.2 177.2 ErrorErrorError % % % 6.13 6.13 5.01 5.01 5.01 3.17 3.17 3.17 4.39 4.39 4.39 4.21 4.21 4.21 3.45 3.45 3.45 3.60 3.60 3.60 S/VS/V S/Vratio ratio ratio 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22

Figure 5. Relation between model resolution and measurement errors. FigureFigure 5. 5. RelationRelation between between model model resolution resolution and and measurement measurement errors. errors.

Figure 6. Fragments of the same object (an Aloe marlothii leaf) depicted as the triangular mesh with Figuredifferent 6. Fragments resolutions ofand the with same the object shortest (an distances Aloe marlothii between leaf) two depicted points. as the triangular mesh with Figure 6. Fragments of the same object (an Aloe marlothii leaf) depicted as the triangular mesh with different resolutions and with the shortest distances between two points. differentOn the resolutionsbasis of the and diagram with the shown shortest in distancesFigure 5, between it can be two said points. that a model with the triangular meshOn resolution the basis ofof the1 mm diagram or less shown yields in errorsFigure lower 5, it can than be 1%.said Suchthat aaccuracy model with is sufficient the triangular for meshmeasuring resolution morphological of 1 mm propertiesor less yields of plants errors wh loweren using than computer 1%. Such models. accuracy Spatial is sufficient triangular for measuring morphological properties of plants when using computer models. Spatial triangular

Sensors 2016, 16, 924 7 of 14

On the basis of the diagram shown in Figure5, it can be said that a model with the triangular mesh resolution of 1 mm or less yields errors lower than 1%. Such accuracy is sufficient for measuring morphological properties of plants when using computer models. Spatial triangular mesh resolution is a parameter of final-stage modeling in the data-processing pipeline. Applying very fine mesh resolutions results in a vast number of triangles, which lengthens the processing time and, in extreme cases, makes calculations impossible due to the limited memory available, especially in 32 bit software. In Dijkstra’s method, the entire triangular mesh dataset is converted to a graph, which requires corresponding memory allocation. Proper mesh resolution is required for accurate measurements. The penultimate row in Table2 lists errors in length calculation between the points, determined with the mesh and the shortest geometric value (blue line shown in Figure2A). This result was taken as the reference measurement. It can be seen that for models 3 to 7 the error is 3% to 5%. For a distance of 180 mm it gives about 7 mm. This error can be treated as the method error, stemming from measurements based on models with triangular mesh as the surface. In phenotyping, it is more important to determine relative values of parameters (growth changes) than their absolute values. In that case, the systematic error resulting from the approximate distance calculation will not play a role. Moreover the S/V ratio, frequently calculated in plant science, remains unchanged regardless of the triangular mesh resolution (Table2). The general assumption in volume calculation is that the model is of closed surface. Measurement accuracy depends on an accurate model and on the measurement procedure. Paulus et al. [26] performed a comprehensive investigation of model accuracy using the DAVID 3D Laserscanner. To enhance this study, a sphere with a radius of 50 mm was used to validate the calculation of volume and surface area. The sphere was triangulated with an angular resolution of 2˝ which gives a distance between vertices of 1.7 mm on the equator. The measured volume and surface area were 523,329 mm3 and 31,408 mm2, which compare well with the calculated volume and surface area of 523,598 mm3 and 31,415 mm2. Thus, the triangulation has negligible error.

3.2. DAVID Laserscanner Scanning Results and Data Processing For specimens smaller than 600 mm, the best device for data acquisition is a structured-light scanner. The authors used the DAVID 3D SLA-1 scanner [17] in their research. A complete model of the plant was obtained by combining 30 partial scans. The Iterative Closest Point (ICP) method was used to combine them. Data acquisition and scan combination were performed with the use of the manufacturer’s software. Figure3 shows the image of the plant ( Welwitschia mirabilis) and its geometric model with the texture visible. After the model in the form of a triangle mesh is obtained, software developed by the authors was used for the calculation of morphological parameters. The software was developed based on the Visualization Toolkit (VTK) library [38,39] commonly used in the computer graphics community. For twisted leaf length estimations such as those for Welwitschia mirabilis leafs a spatial software tool was developed based on the Matlab environment [37].

3.3. Hand-Held and Kinect Scanning Results Owing to plant size limitations when using the structured-light DAVID Laserscanner, the feasibility of working with a hand-held and Kinect scanner was assessed using the Welwitschia mirabilis plant. The device HandyScan EXASCAN produced by Creaform was used [29]. This somewhat older scanner (made in 2009) requires the placement of markers onto the object or its surroundings to aid in positioning the scan head. Current hand-held scanners no longer have this disadvantage and the positioning is performed based on the scanned point cloud. The leaf models are shown in Figure7. The appearance is much worse than for the models obtained with the DAVID Laserscanner (Figure3), as seen in the number of missing fragments. Sensors 2016, 16, 924 8 of 14 Sensors 2016, 16, 924 8 of 14 SensorsSensors 20162016,, 1616,, 924924 88 ofof 1414

Figure 7. Model of a Welwitschia mirabilis leaf obtained with: ( A) A hand-held and ( B) Kinect scanner. FigureFigure 7.7. ModelModel ofof aa WelwitschiaWelwitschia mirabilismirabilis leafleaf obtainedobtained with:with: ((AA)) AA hand-heldhand-held andand ((BB)) KinectKinect scanner.scanner. 3.4. Geometric Models of Large-Size Plants 3.4.3.4. GeometricGeometric ModelsModels ofof Large-SizeLarge-Size PlantsPlants Twenty-nine imagesimages ofof aa specimenspecimen ofof PachypodiumPachypodium lamereilamerei,, photographed with the digitaldigital Twenty-nineTwenty-nine imagesimages ofof aa specimenspecimen ofof PachypodiumPachypodium lamerei,lamerei, photographedphotographed withwith thethe digitaldigital camera Panasonic Lumix DMC-FZ38 in the glasshouse environment, environment, are are presented presented in in the the Figure Figure 88.. cameracamera PanasonicPanasonic LumixLumix DMC-FZ38DMC-FZ38 inin thethe glasshouseglasshouse environment,environment, areare presentedpresented inin thethe FigureFigure 8.8. Image processing and object reconstruction were done with the help of VisualSFM software [40,41] ImageImageImage processing processingprocessing andandand objectobjectobject reconstructionreconstructionreconstruction were were do donedonene withwith thethe help help ofof of VisualSFMVisualSFM VisualSFM softwaresoftware software [40,41][40,41] [40,41 ] and the reconstruction effect can be seen in Figures 9 and 10. andandand the thethe reconstruction reconstructionreconstruction effect effecteffect can cacann be bebe seen seenseen ininin FiguresFiguresFigures9 99 and andand 10 10.10..

FigureFigureFigure 8. 8.8. SeveralSeveralSeveral samplesample imagesimages fromfrom thethe PachypodiumPachypodium lamerei lamereilamerei series seriesseries.. . Figure 8. Several sample images from the Pachypodium lamerei series.

FigureFigureFigure 9. 9.9. Scene SceneScene reconstructed reconstructedreconstructed from from the thethe images imagesimages wi withwithth a a pointpoint point cloud cloud cloud and and and a a a visible visible visible texture.texture. texture.texture.

Sensors 2016, 16, 924 9 of 14 Sensors 2016, 16, 924 9 of 14

FigureFigure 10.10. Model of Pachypodium lamereilamerei..

The point cloud presented in Figure 9 was limited to the area which contains the object to be The point cloud presented in Figure9 was limited to the area which contains the object to be reconstructed. The reconstruction was performed with the Poisson method [42]. The main challenge reconstructed. The reconstruction was performed with the Poisson method [42]. The main challenge in surface reconstruction from a point cloud in photogrammetry is to fit the surface according to the in surface reconstruction from a point cloud in photogrammetry is to fit the surface according to the real object shape in the presence of noise, holes and a sparse point cloud. In the Poisson method, real object shape in the presence of noise, holes and a sparse point cloud. In the Poisson method, because t uses an indicator function defined as 1 for points inside the model and as 0 for points because t uses an indicator function defined as 1 for points inside the model and as 0 for points outside, the reconstructed surface is an extracted isosurface. This result is usually better than that of outside, the reconstructed surface is an extracted isosurface. This result is usually better than that of a Delaunay triangulation. a Delaunay triangulation. Clearly, the model shown in Figure 10 is not suitable for species identification, let alone Clearly, the model shown in Figure 10 is not suitable for species identification, let alone phenotyping. A relatively small number of images influenced the result. To overcome the problem phenotyping. A relatively small number of images influenced the result. To overcome the problem of of large plant digitalization a multi-camera setup as proposed by Nguyen et al. [43] is recommended. large plant digitalization a multi-camera setup as proposed by Nguyen et al. [43] is recommended.

3.5.3.5. Detecting the Effect of Different KindsKinds ofof SurfacesSurfaces onon EffectiveEffective PlantPlant ScanningScanning TheThe largelarge diversity diversity of plantsof plants as regardsas regards their shapetheir shape and type and of surfacetype of issurface sometimes is sometimes not conducive not toconducive obtaining to scan-based obtaining computerscan-based models computer that models serve well that for serve phenotyping. well for phenotyping. In order to assess In order which to surfaceassess which feature surface influences feature this influences problem, this seven problem, plant species seven plant were species scanned were (Figure scanned 11) to (Figure obtain 11) their to computerobtain their models computer (Figure models 12). The (Figure models were12). Th obtainede models with were the DAVID obtained 3D Laserscanner.with the DAVID 3D Laserscanner.It can be claimed that computer models of plants similar to numbers 1 to 4 are usable for measurementIt can be purposes.claimed that Plants computer number models 5 to 7 of are plants not suitable similar forto scanningnumbers with1 to 4 structured are usable light for devicesmeasurement due to purposes. the presence Plants of number thorns and/or 5 to 7 are prickles not suitable which dispersefor scanning the light. with structured Such computer light modelsdevices aredue incomplete to the presence and not of fitthorns for measurement and/or prickles purposes. which disperse Such plants the canlight. be Such scanned computer using models are incomplete and not fit for measurement purposes. Such plants can be scanned using a a triangulation sensor, e.g., Perceptron Scanworks, coupled to a Romer arm. This arrangement allows triangulation sensor, e.g., Perceptron Scanworks, coupled to a Romer arm. This arrangement allows maximum width (140 mm) and depth of field (110 mm), with a minimum point-to-point distance of maximum width (140 mm) and depth of field (110 mm), with a minimum point-to-point distance of 12 micrometers. A measurement arm or robotic scanner can move around the plant to capture data 12 micrometers. A measurement arm or robotic scanner can move around the plant to capture data that would be obscured by single-angle measurements. The scanner price, reportedly $100,000, is, that would be obscured by single-angle measurements. The scanner price, reportedly $100,000, is, however, a drawback. however, a drawback. Before measurements are taken, the model must be corrected to eliminate features not relevant to Before measurements are taken, the model must be corrected to eliminate features not relevant the study, such as a pot. When minor omissions are present, a mesh can be constructed to replace the to the study, such as a pot. When minor omissions are present, a mesh can be constructed to replace missing features using, e.g., Autodesk Meshmixer software. the missing features using, e.g., Autodesk Meshmixer software. For plants that possess scattered thorns or prickles (plants 1 and 4), the computer model is seldom For plants that possess scattered thorns or prickles (plants 1 and 4), the computer model is complete, since the thorns or prickles are visible as empty spaces. The area with empty spots can be seldom complete, since the thorns or prickles are visible as empty spaces. The area with empty spots calculated; the fragment then needs to be filled and recalculated (Figure 13) to achieve results which can be calculated; the fragment then needs to be filled and recalculated (Figure 13) to achieve results are impossible to obtain with standard methods. which are impossible to obtain with standard methods.

Sensors 2016, 16, 924 10 of 14 Sensors 2016, 16, 924 10 of 14 Sensors 2016, 16, 924 10 of 14

Figure 11. Scanned plant species which had computer models created. 1. Aloe marlothii; 2. Figure 11. Scanned plant species which had computer models created. 1. Aloe marlothii; FigureAstrophytum 11. Scanned capricorne; plant 3. Astrophytumspecies which myriostigma had computer; 4. Pilosocereus models created. pachycladus; 1. Aloe 5. Mammillariamarlothii; 2. 2. Astrophytum capricorne; 3. Astrophytum myriostigma; 4. Pilosocereus pachycladus; 5. Mammillaria Astrophytummagnimamma; capricorne; 6. Echinopsis 3. Astrophytum leucantha; 7. myriostigmaLeuchtenbergia; 4. Pilosocereusprincipis. pachycladus; 5. Mammillaria magnimamma; 6. Echinopsis leucantha; 7. Leuchtenbergia principis. magnimamma; 6. Echinopsis leucantha; 7. Leuchtenbergia principis.

Figure 12. Computer models for plants in Figure 11. Figure 12. Computer models for plants in Figure 11. Figure 12. Computer models for plants in Figure 11.

Sensors 2016, 16, 924 11 of 14 Sensors 2016, 16, 924 11 of 14

FigureFigure 13.13. FragmentFragment ofof anan Aloe marlothii leaf with visible pointspoints wherewhere thornsthorns grow:grow: ((AA)) FilledFilled areas;areas; ((BB)) HolesHoles inin placeplace ofof thorns.thorns.

4.4. Discussion Discussion ComputerComputer modelsmodels makemake itit possiblepossible to determinedetermine the area andand volumevolume precisely, whichwhich letslets thethe researcherresearcher calculatecalculate thethe S/VS/V ratio.ratio. A methodmethod ofof approximatingapproximating thethe S/VS/V ratioratio byby usingusing geometricalgeometrical primitivesprimitives existsexists [[44],44], butbut computercomputer modelsmodels cancan calculatecalculate thethe S/VS/V ratioratio moremore accurately.accurately. ThisThis isis especiallyespecially importantimportant forfor thethe ecologicalecological studiesstudies ofof succulentssucculents [[20],20], andand suchsuch anan accurateaccurate methodmethod hashas nevernever beenbeen usedused soso far.far. ThreeThree scannersscanners werewere assessedassessed (DAVID(DAVID Laserscanner,Laserscanner, EXASCAN,EXASCAN, andand Kinect).Kinect). TheThe bestbest resultsresults forfor shapeshape andand completenesscompleteness (no(no missingmissing features,features, e.g.,e.g., asas inin FiguresFigures7 7 and and 10 10)) were were obtained obtained with with the the DAVIDDAVID Laserscanner.Laserscanner. ScanScan time,time, whichwhich takestakes approximatelyapproximately 4040 minmin forfor oneone plantplant (with(with aroundaround 3535 toto 4040 scans),scans), isis thethe onlyonly disadvantagedisadvantage ofof workingworking withwith thisthis device.device. AA detaileddetailed comparisoncomparison of of the the DAVID DAVID 3D 3D and and Kinect Kinect scanners scanners has has been been reported reported [26 [26].]. Phantoms Phantoms of knownof known geometrical geometrical shapes shapes (plane (plane and and sphere), sphere), and an modelsd models of plantsof plants scanned scanned with with the the Perceptron Perceptron v5 coupledv5 coupled to the to Romerthe Romer Infinite Infinite 2.0 arm 2.0 asarm a “groundas a “gro truth”,und truth”, were usedwere toused estimate to estimate measurements measurements errors. Theerrors. error The for error the DAVID for the 3D DAVID device was3D lessdevice then was˘1 mm,less then which ±1 is mm, comparable which withis comparable the manufacturer’s with the specifications,manufacturer’s and specifications, is appropriate and for is phenotypingappropriate for and phenotyping parameterization. and parameterization. DeterminationDetermination of the shortest shortest path path in in a atriangle triangle mesh mesh with with the the use use of ofDijkstra’s Dijkstra’s method method has hasthe theadvantage advantage of correct of correct calculation calculation compared compared with with results results from from commercial commercial software software RapidForm. RapidForm. This Thismethod method is not is notsuitable suitable for fortwisted twisted objects objects such such as asWelwitschiaWelwitschia mirabilis mirabilis leaves,leaves, in in which which the shortestshortest pathpath shouldshould be be determined determined as aas center a center line. Forline. such Fo objectsr such itobjects is necessary it is tonecessary use the more to use sophisticated the more methodsophisticated presented method in Section presented 2.2.2 in. Section 2.2.2. AA robustrobust 3D3D structuredstructured lightlight scanningscanning systemsystem withwith aa setset ofof fivefive stereostereo camerascameras andand softwaresoftware focusedfocused onon wholewhole plant phenotyping has has been been repo reportedrted [45]. [45]. The The use use of of a amotorized motorized table table and and a amultiple-camera multiple-camera set set spread spread over over a plant a plant allows allows us usto toscan scan the the whole whole plant plant in one in one process. process. It is It not is notpossible possible with with the DAVID the DAVID Laserscanner Laserscanner which which is comp is composedosed of one of camera one camera and one and structured one structured light lightsource. source. AA hand-heldhand-held scannerscanner (Artec(Artec S)S) hashas beenbeen reportedreported toto givegive aa goodgood resolutionresolution [[5].5]. However,However, basedbased onon ourour experienceexperience withwith thethe EXASCANEXASCAN device,device, wewe feelfeel otherwise.otherwise. It should,should, however,however, bebe notednoted thatthat EXASCANEXASCAN isis nono longerlonger aa state-of-the-artstate-of-the-art device.device. Advantages of the DAVIDDAVID LaserscannerLaserscanner whenwhen scanningscanning plantsplants areare alsoalso reportedreported [[24],24], andand theythey areare forfor KinectKinect devicedevice asas well;well; however,however, wewe diddid notnot seesee these.these. Scanning results obtained by authors withwith the useuse ofof thethe KinectKinect sensorsensor (Figure(Figure7 7B)B) were were notnot satisfactory.satisfactory. TheThe modelmodel isis incomplete,incomplete, whichwhich makesmakes itit impossibleimpossible toto measuremeasure leafleaf characteristicscharacteristics andand toto comparecompare withwith thethe otherother scanner.scanner. AnotherAnother approachapproach toto KinectKinect sensorsensor datadata processingprocessing hashas beenbeen described [45]. Segmentation of stems and leaves is performed in 2D depth and image color space instead of 3D scanning.

Sensors 2016, 16, 924 12 of 14 described [45]. Segmentation of stems and leaves is performed in 2D depth and image color space instead of 3D scanning. Non-invasive methods of morphological studies of plants are especially important in studies of plants which are endangered and therefore protected by law. With new methods, it is possible to measure their dimensions and consequently also estimate their age, which has not been possible so far. Exact measures of plants with complicated surfaces and leaf structure are now possible, thanks to phenotyping methods.

5. Conclusions This paper presents the possibilities of using optical scanners to scan and prepare computer models of plants for measurement of their morphological parameters. Also presented is the analysis of changes in measurement accuracy depending on the triangular mesh resolution in order to assess which resolution yields sufficient measurement accuracy with a reasonable model complexity. Our work resulted in the development of an original method for leaf length estimation along the line in the middle of the leaf. The method will be used in a study of the Welwitschia mirabilis specimens from Adam Mickiewicz University Botanical Garden in Poznan. It can also be used for other plants with a contorted leaf shape. In the future, such models can be used for ecological studies as well as for creating online applications for authorities involved in plant protection.

Acknowledgments: The authors would like to thank Jerzy Błoszyk and Tomasz Kalinowski from the Department of General Zoology of the Faculty of Biology of the A. Mickiewicz University for their kind help scanning with HandyScan EXASCAN. The authors are grateful to Robert Palmer Jr. (NIDCR, Bethesda), for linguistic correction. The authors wish to thank anonymous reviewers for their valuable comments. This research was supported by funds of the Botanical Garden and the Faculty of Biology of the A. Mickiewicz University in Poznan and Institute of Control and Information Engineering, Poznan University of Technology. Author Contributions: Pawel Drapikowski conducted the measurements, interpreted the data and drafted the manuscript. Dominik Korecki developed a method for leaf length estimation along the line in the middle of the leaf. Ewa Kazimierczak-Grygiel selected and prepared plant material, took part in the experiment performation and text preparation. Justyna Wiland-Szyma´nskacontributed plant material and partially wrote the paper. Conflicts of Interest: The authors declare no conflict of interest.

References

1. Li, L.; Zhang, Q.; Huang, D. A Review of Imaging Techniques for Plant Phenotyping. Sensors 2014, 14, 20078–20111. [CrossRef][PubMed] 2. Chéné, Y.; Rousseau, D.; Lucidarme, P.; Bertheloot, J.; Caffier, V.; Morel, P.; Belin, E.; Chapeau-Blondeau, F. On the use of depth camera for 3D phenotyping of entire plants. Comput. Electron. Agric. 2012, 82, 122–127. [CrossRef] 3. Von Willert, D.J.; Brinckmann, E.; Scheitler, B.; Eller, B.M. Availability of water controls Crassulacean acid metabolism in succulents of the Richtersveld (Namib desert, ). Planta 1985, 164, 44–55. [CrossRef][PubMed] 4. Gaff, D.F. Desiccation tolerant vascular plants of . Oecologia 1977, 31, 95–109. [CrossRef] 5. Schulze, E.-D.; Lange, O.L.; Buschbom, U.; Kappen, L.; Evenari, M. Stomatal responses to changes in humidity in plants growing in the desert. Planta 1972, 108, 259–270. [CrossRef][PubMed]

6. Von Willert, D.J.; Wagner-Douglas, U. Water relations, CO2 exchange, water-use efficiency and growth of Welwitschia mirabilis Hook. fil. in three contrasting habitats of the Namib desert. Bot. Acta 1994, 107, 291–299. [CrossRef] 7. Henschel, J.R.; Seely, M.K. Long-term growth patterns of Welwitschia mirabilis, a long-lived plant of the Namib Desert (including a bibliography). Plant Ecol. 2000, 150, 7–26. [CrossRef] 8. Ehleringer, J. Leaf morphology and reflectance in relation to water and temperature stress. In Adaptation of Plants to Water and High Temperature Stress; John Wiley & Sons, Inc.: New York, NY, USA, 1980; pp. 293–308. 9. Bruce, W.G.G. Medicinal properties in the Aloe. Excelsa 1975, 5, 57–68. Sensors 2016, 16, 924 13 of 14

10. Spickett, A.; Van der Merwe, D.; Matthee, O. The effect of orally administered Aloe marlothii leaves on Boophilus decoloratus tick burdens on cattle. Exp. Appl. Acarol. 2007, 41, 1–2. [CrossRef][PubMed] 11. Grace, O.M.; Simmonds, M.S.J.; Smith, G.F.; van Wyk, A.E. Therapeutic uses of Aloe L. () in southern Africa. J. Ethnopharmacol. 2008, 119, 604–614. [CrossRef] 12. Liége, R.; de Souza, A.; Santana, M.F.S.; de Macedo, E.M.S.; de Brito, E.S.; Correia, R.T.P. Physicochemical, bioactive and functional evaluation of the exotic fruits Opuntia ficus-indica and Pilosocereus pachycladus Ritter from the Brazilian caatinga. J. Food Sci. Technol. 2015, 52, 7329–7336. 13. Goettsch, B.; Hilton-Taylor, C.; Cruz-Piñón, G.; Duffy, J.P.; Frances, A.; Hernández, H.M.; Inger, R.; Pollock, C.; Schipper, J.; Superina, M.; et al. High proportion of cactus species threatened with extinction. Nat. Plants 2015.[CrossRef][PubMed] 14. Convention on International Trade in Endangered Species of Wild Fauna and Flora. Available online: https:// www.cites.org/sites/default/files/eng/disc/CITES-Convention-EN.pdf (accessed on 24 November 2015). 15. The IUCN Red List of Threatened Species Version 2014. 1. Available online: http://www.iucnredlist.org (accessed on 24 November 2015). 16. Rahaman, M.M.; Chen, D.; Gillani, Z.; Klukas, C.; Chen, M. Advanced phenotyping and phenotype data analysis for the study of plant growth and development. Front. Plant Sci. 2015, 10, 1–15. [CrossRef] [PubMed] 17. DAVID 3D Scanning. Available online: http://www.david-3d.com (accessed on 24 November 2015). 18. Suk, J.; Kim, S.; Ryoo, I. Non-contact plant growth measurement method and system based on ubiquitous sensor network technologies. Sensors 2011, 11, 4312–4334. [CrossRef][PubMed] 19. Kicherer, A.; Herzog, K.; Pflanz, M.; Wieland, M.; Rüger, P.; Kecke, S.; Kuhlmann, H.; Töpfer, R. An Automated Field Phenotyping Pipeline for Application in Grapevine Research. Sensors 2015, 15, 4823–4836. [CrossRef][PubMed] 20. Fanourakis, D.; Briese, C.; Max, J.F.J.; Kleinen, S.; Putz, A.; Fiorani, F.; Ulbrich, A.; Schurr, U. Rapid determination of leaf area and plant height by using light curtain arrays in four species with contrasting shoot architecture. Plant Methods 2014, 10, 9. [CrossRef][PubMed] 21. Walter, A.; Liebisch, F.; Hund, A. Plant phenotyping: From bean weighing to image analysis. Plant Methods 2015, 11, 14. [CrossRef][PubMed] 22. Rousseau, D.; Chéné, Y.; Belin, E.; Semaan, G.; Trigui, G.; Boudehri, K.; Franconi, F.; Chapeau-Blondeau, F. Multiscale imaging of plants: Current approaches and challenges. Plant Methods 2015, 11, 6. [CrossRef] [PubMed] 23. Humplík, J.F.; Lazár, D.; Husiˇcková,A.; Spíchal, L. Automated phenotyping of plant shoots using imaging methods for analysis of plant stress responses—A review. Plant Methods 2015, 11, 29. [CrossRef] 24. Mankoff, K.; Russo, T. The Kinect: A low-cost, high-resolution, short-range, 3D camera. Earth Surf. Process. Landf. 2013, 38, 926–936. [CrossRef] 25. Barsanti, S.G.; Remondino, F.; Visintini, D. Photogrammetry and Laser Scanning for Archaeological Site 3D Modeling—Some Critical Issues. In Proceedings of the 2nd Workshop on the New Technologies for Aquileia, Aquileia, Italy, 25 June 2012; pp. B1–B10. 26. Paulus, S.; Behmann, J.; Mahlein, A.-K.; Plümer, L.; Kuhlmann, H. Low-Cost 3D Systems: Suitable Tools for Plant Phenotyping. Sensors 2014, 14, 3001–3018. [CrossRef][PubMed] 27. Kempthorne, D.M.; Barry, M.; Zabkiewicz, J.A.; Young, J. 3D digitisation of plant leaves. ANZIAM J. 2014, 55, C138–C152. 28. Artec 3D Scanners. Available online: http://www.artec3d.com (accessed on 24 November 2015). 29. HandyScan 3D. Available online: http://www.creaform3d.com/en/metrology-solutions/products/ portable-3d-scanner/technical-specifications-handyscan-3d (accessed on 24 November 2015). 30. Perceptron. Available online: http://www.perceptron.com (accessed on 24 November 2015). 31. Hexagon Metrology. Available online: https://hexagonmetrology.us/news-and-events/press-releases/116- romer-inc-unveils-all-new-infinite-20-series-measuring-arm (accessed on 24 November 2015). 32. Portable Living Leaf Area Meter. Available online: http://www.instrumentsmart.com/plant-testing- instrument.html#portable-living-leaf-area-meter (accessed on 24 November 2015). 33. Plant database. Available online: www.tropicos.org (accessed on 12 March 2016). 34. Cormen, T.H.; Leiserson, C.E.; Rivest, R.L.; Stein, C. Introduction to Algorithms, 2nd ed.; MIT Press and McGraw-Hill: Cambridge, MA, USA, 2001; pp. 595–601. Sensors 2016, 16, 924 14 of 14

35. Rapidform. Available online: http://www.rapidform.com/evaluations/rapidform-2006-evaluation (accessed on 24 November 2015). 36. Kimmel, R.; Setian, J.A. Computing geodesic Paths on Manifolds. Proc. Natl. Acad. Sci. USA 1998, 95, 8431–8435. [CrossRef][PubMed] 37. Korecki, D. Software and Hardware Setup for Scanning and Plants Phenotyping. Master’s Thesis, Poznan University of Technology, Poznan, Poland, 2015. (In Polish) 38. Schroeder, W.; Martin, K.; Lorencen, M. The Visualization Toolkit. In An Object-Oriented Approach to 3D Graphics; Prentice-Hall: Upper Saddle River, NJ, USA, 2003. 39. The Visualization Toolkit. Available online: http://www.vtk.org (accessed on 24 November 2015). 40. Wu, C. VisualSFM: A Visual Structure from Motion System. Available online: http://ccwu.me/vsfm (accessed on 24 November 2015). 41. Wu, C. Towards Linear-Time Incremental Structure from Motion. In Proceedings of the 3DV 2013 International Conference on 3D Vision, Seattle, WA, USA, 29 June–1 July 2013; pp. 127–134. 42. Kazhdan, M.; Bolitho, M.; Hoppe, H. Poisson Surface Reconstruction. Eurogr. Symp. Geom. Process. 2006, 61–70. [CrossRef] 43. Nguyen, T.T.; Slaughter, D.C.; Max, N.; Maloof, J.N.; Sinha, N. Structured Light-Based 3D Reconstruction System for Plants. Sensors 2015, 15, 18587–18612. [CrossRef][PubMed] 44. Mauseth, J.D. Theoretical Aspects of Surface-to-Volume Ratios and Water-Storage Capacities of Succulent Shoots. Am. J. Bot. 2000, 87, 1107–1115. [CrossRef][PubMed] 45. Li, D.; Xu, L.; Tan, C.; Goodman, E.D.; Fu, D.; Xin, L. Digitization and Visualization of Greenhouse Tomato Plants in Indoor Environments. Sensors 2015, 15, 4019–4051. [CrossRef][PubMed]

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).