Accurate Plant Modeling based on the Real Light Incidence

J. M. Jurado, J. L. C

ardenas, C. J. Ogayar, L. Ortega and F. R. Feito

Computer Graphics and Geomatics Group, University of Ja

en, Spain

Keywords:

3D Plant Reconstruction, Spectral Reﬂectance, Image Processing, Procedural Modeling.

Abstract:

In this paper, we propose a framework for accurate plant modeling constrained to actual plant-light interaction

along a time-interval. To this end, several plant models have been generated by using data from different

sources such as LiDAR scanning, optical cameras and multispectral sensors. In contrast to previous approaches

that mostly focus on realistic rendering purposes, the main objective of our method is to improve the multi-

view stereo reconstruction of plant structures and the prediction of the growth of existing plants according to

the inﬂuence of real light incidence. Our experimental results are oriented to olive trees, which are formed by

many thin branches and dense foliage. Plant reconstruction is a challenging task due to self-occlusion. Our

approach is based on inverse modeling to generate a parametric model which describes how plants evolve in a

time interval by considering the surrounding environment. A multispectral sensor has been used to characterize

input plant models from reﬂectance values for each narrow-band. We propose the fusion of heterogeneous data

to achieve a more accurate modeling of plant structure and the prediction of the branching fate.

1 INTRODUCTION

Realistic plant modeling is a well known topic of

research in Computer Graphics. Its applications

are mainly visualization and virtual reality. Other

disciplines have also included plant reconstruction

as a research objective, such as remote sensing

(Prusinkiewicz, 2004) and biology (Omasa et al.,

2006). Many plant modeling approaches have been

developed over the last years. However, the dynamic

plant behaviour, by considering the surrounding envi-

ronment, remains a non-trivial and challenging task.

Procedural modeling approaches can efﬁciently

synthesize the branching structure of existing real-

world plant from a set of rule-based system (Bene

et al., 2011) and (Guo et al., 2018). In general,

these methods are aimed to obtain visually accept-

able results for rendering and simulation purposes,

and therefore they are not directly oriented for 3D

modeling real-world vegetation.

Geometry-based methods may also accurately re-

construct the skeletal structure of a tree, but the fo-

liage is difﬁcult to recreate (Xu et al., 2007; Livny

et al., 2010). One of the most promising image-based

methods is Structure-from-Motion (SfM), which is

widely used to generate 3D point clouds from mul-

tiple overlapping images (Lou et al., 2014). However,

this method arises some limitations, and the recon-

struction of complex plant structures is prone to er-

rors. On the other hand, other approaches may effec-

tively reconstruct real plants using an inverse proce-

dural method (Stava et al., 2010). In addition to ge-

ometrical data, some methods also regard additional

environmental factors, such as light incidence (Stava

et al., 2014), which inﬂuences the growth of the plant.

However, these estimations are based on a probabilis-

tic model, and not measured from the actual plant.

In this paper, we propose a several improvements

for plant modeling constrained by the real plant-light

interaction along a time interval. We aim to fuse

the plant static reconstruction and inverse procedu-

ral modeling for plant growth prediction. In contrast

to previous approaches that mostly focus on realistic

rendering purposes, the main objective of our method

is to monitor and predict the growth of real plants.

We have tested our approach with olive trees, which

are formed by a complex crown structure with many

self-hidden branches and leaves. An inverse modeling

process is applied for generating the parametric model

to describe several botanic features. To this end, mul-

tispectral images are used to extract reﬂectance in-

dices for each narrow-band to estimate the plant vigor

and predict its next growth. A semantic classiﬁcation

of the plant shape is carried out by considering how

plants reﬂect the light energy.

The paper is organized as follows. Firstly, a re-

view of previous work on plant modeling is presented

(Section 2) and we provide the general overview of

360

Jurado, J., Cárdenas, J., Ogayar, C., Ortega, L. and Feito, F.

Accurate Plant Modeling based on the Real Light Incidence.

DOI: 10.5220/0007686803600366

In Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2019), pages 360-366

ISBN: 978-989-758-354-4

our method (Section 3). In the following, experimen-

tal results are shown which are obtained from LiDAR

and multiple images (Section 4) and the fusion of

multispectral data to estimate the real light incidence

to the plant model (Section 5). Finally, we discuss

the contributions of the paper and the main topics for

further research (Section 6).

2 RELATED WORK

Plant modeling is a classic topic of research in Com-

puter Graphics. However, the main efforts have been

made for visualization and virtual reality purposes. In

this work, we apply plant modeling for the monitor-

ing and prediction of the evolution of actual plants

and bushes from extensive plantations. According

to the literature, plant modeling approaches may be

mainly classiﬁed into three categories: reconstruction

from existing real-world plants, interactive modeling

and procedural or rule-based systems, such as fractals

(Oppenheimer, 1986) and L-Systems (Prusinkiewicz,

1986). Neither geometry-based nor interactive meth-

ods for tree shape reconstruction takes into account

any environmental effects. On the other hand, proce-

dural methods are capable to generate dynamic plant

models whose evolution can be affected by chang-

ing conditions of the environment (Guo et al., 2018).

There are two main methods which may be used for

modeling existing real-world plants: plant static re-

construction and procedural modeling.

Plant Static Reconstruction. Geometry-based meth-

ods mainly depend on the input data, e.g., the image

quality, environmental light during the capture pro-

cess, the shape and texture of the target plant. In

this way, 3D laser scanning can precisely reconstruct

branching plants (Omasa et al., 2006) and can be used

for the automatic reconstruction of the plant skeleton

without overlapped trees segmentation (Livny et al.,

2010). However, this technique implies some draw-

backs as the sensitivity to occlusion, high device cost,

and tedious application in many complex scenes, es-

pecially terrestrial laser scanning (TLS). As a solution

of this problem the airborne LiDAR means a more

efﬁcient solution for scanning extensive plantations

with an ever increasing precision.

On the other hand, multi-view stereo reconstruc-

tion is another category of methods which can be suc-

cessfully used for tree reconstruction. It is based on a

feature-matching process between multitude overlap-

ping images. In this scope, the SfM method is com-

monly applied to generate a 3D point cloud of a plant

by using several images (Quan et al., 2006). This

technique overcomes some problems mentioned be-

fore, although it may not perform efﬁciently on com-

plex and heterogeneous surfaces where the detection

of key features is more complex. Moreover, this algo-

rithm is based on the scale-invariant feature transform

(SIFT) which may cause errors for the reconstruc-

tion of complex plant models (Lowe, 2004). Conse-

quently, multi-view based realistic tree modeling with

botanical features still pose several limitations. How-

ever, these problems might be partially solved by gen-

erating depth maps for each view, and fusing them

into a dense point cloud reconstruction (Guo et al.,

2018).

Procedural Modeling. Early approaches of plant

modeling focus on the generation of repetitive pat-

terns as fractals or L-Systems (Deussen et al., 1998).

Instead of modeling directly from generative rules,

guided procedural methods have been proposed to

simulate the plant structure from an accurate point

cloud based representation of its branching struc-

ture. Bennes et al. (Bene

s et al., 2011) introduce

guided procedural modeling with several geometric

constraints. Nowadays, the reconstruction from point

clouds has received considerable attention (Berger

et al., 2017). In this way, procedural models can

be transformed into certain shapes depending on the

point cloud input that are acquired with static plant

modeling. Modeling trees according to desired shapes

is important for many applications. However, proce-

dural methods have some drawbacks due to the com-

plexity to determine a valid set of input parameters,

such as branching angle, apical and lateral light ef-

fects, pruning factors, growth rate, etc. These features

are very important for the structure of young trees but

they become less relevant for mature plants. Stava et

al. (Stava et al., 2014) proposed an inverse procedu-

ral modeling method for trees, based on a novel para-

metric model that uses Monte Carlo Markov Chains

(MCMC) for calculating the optimal set of procedu-

ral parameters. However, all these approaches gen-

erate static branching structures rather than modeling

dynamic plant behavior.

Procedural methods are capable of generating tree

models by explicitly considering the environment.

Recent approaches are focused on providing more ef-

ﬁcient ways for plant-environment interaction model-

ing. Palubicki et al. (Palubicki et al., 2009) optimize

branch distribution with space colonization method

(Runions et al., 2007) and local competition for light

resources. In each iteration, the space surrounding of

each bud and the optimal direction of shoot growth

are calculated. Both values are needed to create a tree

structure by simulating real environmental properties.

Later, Lei Yi et al. (Yi et al., 2015) demonstrates the

sensitivity of branching distribution to ambient light

Accurate Plant Modeling based on the Real Light Incidence

361

Figure 1: Overview of the method for olive tree reconstruction from the fusion of multispectral data.

and how each branch inﬂuences to others.

In addition, an important research topic in Ecol-

ogy and Biology focuses on plant phenotyping and

measuring leaf chlorophyll concentration, in order to

study each growing stage of plants from image-based

remote sensing (Moran et al., 1997). In contrast

to the simply visual detection of diseases, narrow-

band sensors are commonly used to measure plant

changes considering how the vegetation interacts with

the ambient light of the surrounding environment

(Candiago et al., 2015). Unfavorable plant grow-

ing results in morphological, physiological and bio-

chemical changes that are determined by the quan-

tity of absorbed or reﬂected light. Leaf spectral re-

ﬂectance provides multiple key features for assessing

plant health; however, leaves typically have a low re-

ﬂectance in the visible spectral due to the high chloro-

phyll absorption (Pe

nuelas and Filella, 1998). Mul-

tispectral sensors capture several spectral bands to

detect many properties, such as drought stress, heat

stress, nutrient content and plant biomass. Current ap-

proaches mostly provide a 2D-based analysis for plant

diseases detection (Thomas et al., 2018), image-based

vegetation segmentation (Suh et al., 2018), plant

phenotyping through ground-based sensors (Sankaran

et al., 2010) and LiDAR and hyperspectral remotely

sensed data (Hakkenberg et al., 2018). Although

many of these methods have certain capabilities to de-

scribe growth behaviour, it is quite difﬁcult to acquire

several plant traits which are not directly visible from

remote sensing imagery. This issue might be over-

come through a faithful 3D modeling of the real plant

for a readily comprehensive assessment of complex

plant features (Klodt and Cremers, 2014).

3 OVERVIEW

In this paper, we propose a method for dynamic plant

modeling constrained to actual plant-light interaction

along a time-interval. The incident light is obtained

from several spectral bands and used to improve the

plant static modeling. The main goal is to generate

accurate plant models for monitoring extensive plan-

tations using several 3D capture techniques, working

at different levels of detail. In this work, we focus

on the reconstruction and characterization of a sin-

gle tree model. An inverse modeling process gen-

erates the parametric model from multispectral data

which provide several features to describe the plant

vigor such as the reﬂectance index, chlorophyll ﬂuo-

rescence or leaves temperature. The generation of the

tree structure by procedural methods is constrained by

several constraints related to the plant status. We also

consider time-lapse data for analyzing the evolution

of the plant along a period. In addition, multispectral

data may be the support for performing reasonably re-

alistic predictions.

The ﬂow diagram of our framework illustrates

the overall process (Figure 1). The ﬁrst step of our

method is the accurate point cloud reconstruction of

each plant at a detailed level. The ﬁrst step of our

method is the plant static modeling of the olive tree

from multi-view stereo reconstruction method and

terrestrial LiDAR scanning. Both geometry models

may be the input for our framework which is based on

multispectral image projection from multiple views.

Data from narrow-band views are used to carry out an

early segmentation of the point cloud, in order to sim-

plify the number of features used in the reconstruc-

GRAPP 2019 - 14th International Conference on Computer Graphics Theory and Applications

362

tion. In the following, an inverse procedural modeling

is performed on the point cloud to overcome recon-

structing issues as self-hidden branches and leaves of

the plant. To this end, for each view a reﬂectance

map is computed for the automatic model segmen-

tation. This technique is useful to determine several

constraints for the guided structure generation. We in-

tegrate this into a rule-based growing system, that also

uses a variation of space colonization (Runions et al.,

2007) and shadow-propagation (Palubicki, 2012). As

our method is based on the change in light distribution

from several spectral bands, the illumination of differ-

ent part of the tree is calculated and used to improve

the data-driven reconstruction. In addition, the spatio-

temporal analysis of the plant behavior provides us

real historic data for the growth prediction.

4 POINT CLOUD

RECONSTRUCTION

One of the main steps of our solution consists of

obtaining three-dimensional information from real

plants to complete it with data obtained from mul-

tispectral sensors. For this purpose, several known

techniques might be applied to generate 3D models

of existing real-world plants. This process can be

carried out for each plant or using batches, depend-

ing on the technology used for capturing the data.

We mainly used two methods: Light Detection and

Ranging (LiDAR), and the image-based reconstruc-

tion. These methods produce distinct results, that is,

point clouds with different attributes and spatial distri-

bution. However, both of them offer spatial data with

enough precision to be used in the following steps of

the modeling.

LiDAR. This 3D scanning technique determines the

distance to a point in space by timing the round trip

time of a light pulse typically ﬁred by a laser diode.

The time that passes until the reﬂected light is cap-

tured by a detector is timed. Visible (green) or in-

visible (near infrared or NIR) lasers are usually used.

Typical time-of-ﬂight laser scanners can measure the

distance of several hundred points per second (high-

density clouds). Airborne LiDAR can scan a wide

area of terrain in a short period of time, and this is

the main reason for using it with the aim of monitor-

ing the temporal evolution of entire plantations. The

main drawback is that 3D scanning is a process prone

to occlusion, especially when capturing plants (Figure

2). However, scanned plants have a higher resolution

than image-based models, hence, more accurate guide

is generated for the inverse modeling.

In order to obtain the skeleton of a tree from the

Figure 2: Point cloud from terrestrial LiDAR scanner.

scanned point cloud, we follow a similar approach as

the ﬁrst step of the algorithm presented in (Xu et al.,

2007). Due to the typical incompleteness of the point

clouds produced by LiDAR, additional branches have

to be synthesized to complete the tree topology, es-

pecially the crown. We use cluster edges in a span-

ning graph to reconstruct the tree skeleton. However,

leaves are not randomly added to the ﬁne branches as

presented in (Xu et al., 2007). Instead, we rely only

on the scanned data, because our method is targeted

to the monitoring of actual leaves. Our experiments

were mainly targeted to olive trees. This type of plant

is formed by many branches and a very dense crown

with multitude of leaves of different sizes. This ag-

gravates the occlusion of the inner part of the tree.

Multi-view Stereo Reconstruction. This method

supposes a simpliﬁcation of the technique of stereo

vision for the case of a single camera that takes im-

ages from different positions. Computer vision tech-

niques are used to calculate matching points, since

the positions of the camera do not have scale and

orientation information in relation to the object to

be scanned. This low cost technique is based on

an adjustment procedure that uses a database of fea-

tures. Those features are automatically extracted from

a set of multiple overlapping images. We have used

a combination of SfM and Patch-based MultiView

Stereo (PMVS) to obtain dense point clouds from

trees. The data obtained with this method can com-

plete the data obtained with LiDAR when necessary,

especially with occluded zones at low heights. SfM is

a technique for producing dense point clouds based on

a feature matching process. However, it does not pro-

duce accurate results on plant images with repetitive

or similar feature properties (Guo et al., 2018). For

this reason, we only use the part of SfM that estimates

the stereo camera position for each of the shots. Then,

PMVS produces a point cloud based on the features

found, taking into account the visibility constraints.

This combination of SfM and PMVS produces better

results, but it still has problems with occlusions and

the performance does not scale well. In order to im-

prove the process, a novelty of our method consists

of using images from the multispectral camera for an

Accurate Plant Modeling based on the Real Light Incidence

363

early segmentation of points, based on multispectral

analysis. This step allows us to extract only the points

corresponding to trunks and branches, omitting all

the leaves, which are precisely the part of the plant

that causes all the feature matching issues. After the

views matching, all points are used for the resulting

dense point cloud (Figure 3). This process greatly

increases the precision of the multi-view reconstruc-

tion, especially with overlapped trees. Later, we use

a point clustering process for the inverted modeling

of the tree skeleton. We also plan to incorporate the

method presented in (Guo et al., 2018). It uses a depth

map based reconstruction algorithm. For each view,

a dense depth map is computed. Then, all maps are

projected into a single point cloud model by consid-

ering visibility. Our contribution at this point involves

the addition of multispectral information to the depth

maps. It provides the removal of all superﬂuous infor-

mation from each map, increasing the performance of

the process.

Figure 3: Point cloud from Structure-from-Motion.

5 PLANT-LIGHT INTERACTION

The surrounding environment plays a key role for

the growing process of trees. The crown shape and

branching structure are directly determined by several

environmental effects. The light energy is considered

as the main factor for branch shooting and it is calcu-

lated by sampling the environmental space. For this

purpose, plant models are decomposed into a grid of

voxels. In this way, a coarse estimation of the inter-

action of light with each bud is computed (Palubicki

et al., 2009). In general, the exposure of each bud to

light is an estimation based on the plant space classi-

ﬁcation into a grid of voxels. In this way, the shadow

propagation method is applied to calculate the bud

fate by selecting the adjacent voxel with the lowest

shadow value. In terms of realistic plant behaviour

according to the light sensitivity, multispectral sen-

sors may be used to capture the reﬂected energy in

various spectral bands.

In this work we have targeted the olive tree as the

main objective. It has been observed by four spectral

bands: Green (530nm-570nm), NIR (770nm-810nm),

Figure 4: Narrow-band channels for each capture.

Red (640nm-680nm) and REG (730nm-840nm) (Fig-

ure 4). For each one, the inﬂuence of the sunlight

is studied for multiple purposes: (1) the 3D recon-

struction of the tree structure from multi-views, (2)

the plant space decomposition and (3) the prediction

of natural growth process. To this end, our method

uses as input either some geometric tree models by

LiDAR scans or multiple overlapping images. Once

the 3D point cloud has been extracted, reﬂectance

maps are computed for each band. Unlike previous

works, which are based on a coarse light propagation,

we calculate the actual light reﬂectance value (LRV)

for each plant from multispectral image processing.

Following, a backward projection is carried out for

point cloud labeling (Figure 5). For this purpose, the

ﬁsheye distortion model of narrow-band images is ap-

plied for mapping the distorted pixel coordinates to

the 3D points.

Figure 5: The multispectral images are back-projected to

the input point-cloud.

Plant Segmentation. Since the input point cloud may

not be completed and many branches have not been

reconstructed due to the occlusion with each others,

a novel 3D classiﬁcation method is proposed. It is

based on the inclusion of reﬂectance values to com-

plete the branching structure and the foliage. As

plants become healthier, the intensity of reﬂectance

increases in the NIR band and decreases in the Red

band, which is the physical basis for most vegetation

indices. NDVI (Normalized Difference Vegetation

Index) value arguably indicates the plant vigor and

is also effective for distinguishing vegetation from

branches, trunk and soil. As Figure 6 shows, the olive

GRAPP 2019 - 14th International Conference on Computer Graphics Theory and Applications

364

tree is subdivided in two classes which correspond

to the foliage and the ramiﬁcation of branches. This

method provides the capability to reduce the noise of

the 3D model which is produced by the reconstruction

of leaves and narrow branches. Therefore, several im-

provements are in progress aimed for plant modeling

from multi-view images by considering the search of

key points speciﬁcally in branches and trunk.

Figure 6: Plant model segmentation.

Plant Space Decomposition. In addition, instead of

performing a voxelization of the plant space, we carry

out a semantic subdivision of the model for studying

shadow propagation. According to previous plant dis-

semination, a 3D clustering method is proposed for

the inverse modeling of the tree structure (Figure 7).

We focus on the classiﬁcation of several point groups

to identify the plant skeleton and improve the guided

modeling of the real tree shape. In the image below,

the soil, trunk and main branches are joined to the

same cluster. This group shares similar reﬂectance

values, ranging from 0 to 0.4 and several soft geo-

metric constraints such as distance, normal direction,

cluster volume, etc. Moreover, the plant crown is sub-

divided into six clusters to estimate the canopy den-

sity and the spanning for each branch.

Figure 7: Semantic classiﬁcation of the plant shape.

Procedural Modeling. The inﬂuence of the sur-

rounding environment should be considered for the

prediction of the plant growth. In this context, the pro-

cedural modeling is capable to simulate the adaptivity

of trees by considering several environmental effects.

In general, the space colonization and the shadow

propagation are the most used methods. However, un-

like previous works, which are based on a coarse es-

timate of bud exposure to calculate the optimal grow-

ing direction, our method takes the real light inci-

dence and the photosynthetic activity of plants from

multispectral imagery. Consequently, it provides sev-

eral parameters to predict the next growth stage ac-

cording to the reﬂectance index which may determine

the pruning factor, phototropism and gravitropism,

branching angle, etc. Our hypothesis is based on es-

timating the growth rate for each region of the plant

due to some parts grow faster than others.

6 CONCLUSIONS AND FUTURE

WORK

We have introduced main contributions of our re-

search in progress. A novel framework is provided

for plant modeling according to the plant-light inter-

action. The key feature of our method is the com-

bination of geometry plant models to semantic data

from multispectral imagery. In summary, plant mod-

eling without the inﬂuence of the surrounding envi-

ronment results to static and unrealistic tree repre-

sentations. This work is focused on the olive tree

structure and their interactions with the environmen-

tal light. In this paper, several techniques for plant

modeling and novel methods based on spatial clas-

siﬁcation have been applied to estimate the growing

shoot from the light-based plant behaviour. To this

end, a plant segmentation is carried out to dissemi-

nate the branching structure to the foliage. Then, a

semantic and volumetric plant space decomposition

is proposed for spatial pooling by several features in

common. Moreover, a historic set of spectral images

have been acquired to predict next growing stages of

target trees from the reﬂectance absorption.

Several open problems may be approached for fu-

ture work. For the monitoring of extensive planta-

tions, it is not feasible to store a dense point cloud

for every plant or bush. First, we would general-

ize our method to model larger scale scenes, such as

real ecosystem of even a forest. Second, since the

plant space decomposition, we focus on generating a

rule-based system to reconstruct self-hidden geome-

try from previous knowledge of how each plant re-

gions interact with the environment. Finally, the de-

velopment of a robust predictive model based on mul-

tispectral or hiperspectral data to simulate the grow-

ing process under different environmental effects.

Accurate Plant Modeling based on the Real Light Incidence

365

ACKNOWLEDGEMENTS

This work has been partially supported by the Min-

isterio de Econom

ıa y Competitividad and the Euro-

pean Union (via ERDF funds) through the research

project (TIN2017-84968-R).

REFERENCES

Bene

s, B., Stava, O., M

ech, R., and Miller, G. (2011).

Guided procedural modeling. In Computer graphics

forum, volume 30, pages 325–334. Wiley Online Li-

brary.

Berger, M., Tagliasacchi, A., Seversky, L. M., Alliez, P.,

Guennebaud, G., Levine, J. A., Sharf, A., and Silva,

C. T. (2017). A survey of surface reconstruction from

point clouds. In Computer Graphics Forum, vol-

ume 36, pages 301–329. Wiley Online Library.

Candiago, S., Remondino, F., De Giglio, M., Dubbini,

M., and Gattelli, M. (2015). Evaluating multispec-

tral images and vegetation indices for precision farm-

ing applications from uav images. Remote Sensing,

7(4):4026–4047.

Deussen, O., Hanrahan, P., Lintermann, B., M

ech, R., Pharr,

M., and Prusinkiewicz, P. (1998). Realistic modeling

and rendering of plant ecosystems. In Proceedings of

the 25th Annual Conference on Computer Graphics

and Interactive Techniques, SIGGRAPH ’98, pages

275–286, New York, NY, USA. ACM.

Guo, J., Xu, S., Yan, D.-M., Cheng, Z., Jaeger, M., and

Zhang, X. (2018). Realistic procedural plant model-

ing from multiple view images. IEEE transactions on

visualization and computer graphics.

Hakkenberg, C., Peet, R., Urban, D., and Song, C. (2018).

Modeling plant composition as community continua

in a forest landscape with lidar and hyperspectral re-

mote sensing. Ecological Applications, 28(1):177–

190.

Klodt, M. and Cremers, D. (2014). High-resolution plant

shape measurements from multi-view stereo recon-

struction. In European Conference on Computer Vi-

sion, pages 174–184. Springer.

Livny, Y., Yan, F., Olson, M., Chen, B., Zhang, H., and

El-Sana, J. (2010). Automatic reconstruction of tree

skeletal structures from point clouds. ACM Trans.

Graph., 29(6):151:1–151:8.

Lou, L., Liu, Y., Han, J., and Doonan, J. H. (2014). Ac-

curate multi-view stereo 3d reconstruction for cost-

effective plant phenotyping. pages 349–356.

Lowe, D. G. (2004). Distinctive image features from scale-

invariant keypoints. International journal of computer

vision, 60(2):91–110.

Moran, M. S., Inoue, Y., and Barnes, E. (1997). Opportuni-

ties and limitations for image-based remote sensing in

precision crop management. Remote sensing of Envi-

ronment, 61(3):319–346.

Omasa, K., Hosoi, F., and Konishi, A. (2006). 3d li-

dar imaging for detecting and understanding plant re-

sponses and canopy structure. Journal of experimental

botany, 58(4):881–898.

Oppenheimer, P. E. (1986). Real time design and anima-

tion of fractal plants and trees. In ACM SiGGRAPH

Computer Graphics, volume 20, pages 55–64. ACM.

Palubicki, W. (2012). Fuzzy Plant Modeling with OpenGL:

Novel Approaches in Simulating Phototropism and

Environmental Conditions. AV Akademikerverlag.

Palubicki, W., Horel, K., Longay, S., Runions, A., Lane,

B., M

ech, R., and Prusinkiewicz, P. (2009). Self-

organizing tree models for image synthesis. In ACM

Transactions on Graphics (TOG), volume 28, page 58.

ACM.

nuelas, J. and Filella, I. (1998). Visible and near-infrared

reﬂectance techniques for diagnosing plant physiolog-

ical status. Trends in plant science, 3(4):151–156.

Prusinkiewicz, P. (1986). Graphical applications of l-

systems. In Proceedings of graphics interface, vol-

ume 86, pages 247–253.

Prusinkiewicz, P. (2004). Modeling plant growth and devel-

opment. Current Opinion in Plant Biology, 7(1):79 –

83.

Quan, L., Tan, P., Zeng, G., Yuan, L., Wang, J., and Kang,

S. B. (2006). Image-based plant modeling. In ACM

Transactions on Graphics (TOG), volume 25, pages

599–604. ACM.

Runions, A., Lane, B., and Prusinkiewicz, P. (2007). Mod-

eling trees with a space colonization algorithm. NPH,

7:63–70.

Sankaran, S., Mishra, A., Ehsani, R., and Davis, C. (2010).

A review of advanced techniques for detecting plant

diseases. Computers and Electronics in Agriculture,

72(1):1–13.

Stava, O., Benes, B., Mech, R., Aliaga, D. G., and Kristof,

P. (2010). Inverse procedural modeling by automatic

generation of l-systems. Comput. Graph. Forum,

29(2):665–674.

Stava, O., Pirk, S., Kratt, J., Chen, B., M

ech, R., Deussen,

O., and Benes, B. (2014). Inverse procedural mod-

elling of trees. In Computer Graphics Forum, vol-

ume 33, pages 118–131. Wiley Online Library.

Suh, H. K., Hofstee, J. W., and van Henten, E. J. (2018). Im-

proved vegetation segmentation with ground shadow

removal using an hdr camera. Precision Agriculture,

19(2):218–237.

Thomas, S., Kuska, M. T., Bohnenkamp, D., Brugger,

A., Alisaac, E., Wahabzada, M., Behmann, J., and

Mahlein, A.-K. (2018). Beneﬁts of hyperspectral

imaging for plant disease detection and plant protec-

tion: a technical perspective. Journal of Plant Dis-

eases and Protection, 125(1):5–20.

Xu, H., Gossett, N., and Chen, B. (2007). Knowledge and

heuristic-based modeling of laser-scanned trees. ACM

Trans. Graph., 26(4).

Yi, L., Li, H., Guo, J., Deussen, O., and Zhang, X. (2015).

Light-Guided Tree Modeling of Diverse Biomorphs.

In Stam, J., Mitra, N. J., and Xu, K., editors, Pa-

ciﬁc Graphics Short Papers. The Eurographics Asso-

ciation.

GRAPP 2019 - 14th International Conference on Computer Graphics Theory and Applications

366