https://doi.org/10.31449/inf.v48i15.6294  Informatica 48 (2024) 179-190   179 
 
3D Reconstruction of Grassland Landforms Using Intelligent Robot 
Vision and Numerical Simulation 
 
Min Zhong
1,2
, Zhanxue Zhou
2*
 
1
School of Architecture and Engineering, Jiangsu Open University, Nanjing, Jiangsu, 210036, China 
2
Civil Engineering College, Hebei University of Architecture, Zhangjiakou, Hebei, 075000, China 
E-mail: zzx1493@hebiace.edu.cn 
*
Corresponding author 
 
Keywords: characteristic numerical simulation, grassland landform, intelligent robot vision system, three-dimensional 
reconstruction  
 
Recieved: May 29, 2024 
Grassland is an important component in the natural ecosystem. Geological detection and numerical 
analysis of grassland landforms are the main ways to protect and develop grassland ecological 
environment. However, the traditional grassland geological survey is not effective. The displayed 
grassland landforms are not intuitive and the numerical fitting accuracy is low, seriously affecting the 
work efficiency of grassland landform survey. With the development of machine vision, cloud 
computing and other technologies, the intelligent robot vision system has stable spatial positioning 
and construction capabilities. Using the intelligent robot vision system, the grassland landform was 
reconstructed in 3D (three-dimension) and compared with the traditional grassland landform 
construction. The experimental results showed that the accuracy of the 3D modeling based on the 
intelligent robot vision system and the traditional 3D reconstruction in the sixth test of the square 
grassland with a side of 2km was 96% and 72% respectively. In a square grassland area with a side 
length of 2km, the average matching degrees of the two 3D modeling feature data were 89.3% and 
73.3% respectively. The study conclude that 3D reconstruction of grassland landforms based on the 
intelligent robot vision system can greatly shorten the time of 3D reconstruction and improve the 
efficiency of 3D reconstruction of grassland landforms. 
Povzetek: Narejena je 3D rekonstrukcijo travniških reliefov z uporabo inteligentnega robotskega 
vidnega sistema, kar izboljša točnost in učinkovitost geoloških analiz v primerjavi s tradicionalnimi 
metodami. 
 
1  Introduction 
With the continuous development of industrial 
technology, the quality and scale of the natural ecological 
environment are declining year by year. Ecological 
environment detection is a data description of the quality 
of the ecological environment. Among them, grassland 
landforms are one of the main ecological environments. 
The detection of grassland landforms can help protect the 
grassland environment, maintain the diversity of 
grassland species, and effectively predict and analyze 
grassland geological changes. In the traditional grassland 
landform detection process, the landform analysis or 
simple three-dimensional analysis is performed in two-
dimensional forms such as drawing a plan, a broken line, 
and a profile. However, the data cannot fully display the 
complex grassland geological data, wasting a lot of 
detection time, and the obtained grassland landform 
detection effect is very poor. With the development of 
intelligent technology, robot vision and other 
technologies are combined with multiple fields. 
Computer vision system technology has super-strong 
spatial 3D construction capabilities, which can accurately  
 
reconstruct grassland landforms in 3D by analyzing the 
characteristic data of grassland landforms. Applying the 
intelligent computer vision system to the geological 
survey of grassland landforms can establish an accurate 
three-dimensional model, improve the accuracy of 
grassland feature data analysis, and improve the work 
efficiency of grassland landform geological survey. 
Therefore, this paper has research significance. 
Geological detection of grassland landforms is conducive 
to maintaining the grassland ecological environment. 
Many people have carried out three-dimensional 
reconstruction and characteristic data simulation of 
grassland landforms. Among them, Huang Y's research 
showed that three-dimensional construction of grassland 
landforms was carried out, and the structural composition 
of grassland landforms was analyzed, which improved 
the targeted protection of grassland environment [1]. Sun 
R conducted three-dimensional modeling and analysis of 
a variety of grassland types, and the species of grassland 
was gradually decreasing [2]. Zhang J M carried out 
three-dimensional modeling of grassland geological soil, 
and obtained the composition of grassland soil through 
visual analysis [3]. Xu Y's research pointed out that 3D 

180    Informatica 48 (2024) 179-190   M. Zhong et al. 
 
reconstruction of grassland landforms can improve the 
accuracy of geological detection and improve the 
detection efficiency [4]. According to Sui D's research, 
the analysis accuracy of the characteristic data of the 
three-dimensional construction of grassland landforms 
was higher than that of the traditional two-dimensional 
plane construction [5]. Although the 3D reconstruction of 
grassland landforms can accurately analyze the grassland 
geological structure, it lacks intelligent technology to 
assist. 
The intelligent robot vision system can realize accurate 
spatial positioning analysis, and apply the intelligent 
robot vision system to the three-dimensional construction 
of grassland landforms. Among them, Zhang Y achieved 
3D reconstruction of grassland landforms through 
binocular vision technology, which improved the 
accuracy of 3D construction [6]. Cheng C's research 
showed that intelligent robot vision can denoise 3D 
reconstruction, improving the stability of 3D construction 
[7]. Park S proposed a robot that runs two cameras 
simultaneously for 3D construction, and an accurate 3D 
model can be obtained by analyzing the eigenvalues [8]. 
Wang J's research pointed out that the intelligent 
computer vision system can effectively construct three-
dimensional grassland landforms by using rectangular 
feature detection [9]. Although intelligent robotic vision 
systems can improve the accuracy of 3D construction, 
there is a lack of comparison with traditional grassland 
landform surveys. 
Surveying grassland landforms is an effective 
supervision measure for grassland ecological 
environment, which can analyze the complex geological 
structure of grassland landforms. Using the intelligent 
robot vision system, the grassland landform was 
reconstructed in 3D, and its characteristic values were 
simulated and analyzed. Compared with the traditional 
3D modeling of grassland landforms, the grassland 
landforms based on the intelligent robot vision system 
can effectively improve the accuracy of data construction 
[10]. 
2  Related works 
In recent years, technological advances have had a 
substantial impact on a variety of sectors, especially 
monitoring of the environment and landform assessment. 
Numerous studies have investigated various approaches 
for assessing soil parameters, detecting landforms, and 
reconstructing three-dimensional models of natural and 
artificial structures. This relevant work to emphasize 
advances and constraints of existing methodologies, 
emphasizing the requirement for creative approaches in 
the area of 3D reconstruction of grassland landforms. 
 
 
 
Table 1: Overview of related research in environmental tracking and landform assessment 
Reference No Objective Methodology Result Limitations 
[11] To assess the 
chemical 
characteristics of 
reclaimed soil in an 
arid grassland 
mining location. 
Variance, 
correlation, and 
principal 
component analysis 
are used to evaluate 
soil characteristics 
and spatial-
temporal variations. 
Reclamation enhanced 
soil chemical 
characteristics, and 
surface soil health 
enhanced with time. 
Constrained to a 
particular mining 
location; may not 
apply to other 
places. 
[12] To create a semi-
automated object-
based image 
analysis (OBIA) 
method for desert 
landform discovery 
and modeling. 
Using DEM and 
spectral data, a 
multi-resolution 
segmentation 
method and object-
based rule sets are 
applied. 
Desert landforms were 
detected and classified 
with excellent 
accuracy (96.21%-
95.05%). 
Concentrated on 
desert landscapes, 
not directly 
relevant to 
grasslands. 
[13] Determine 3D 
macropore networks 
in forest soils and 
compare them to 
agriculture and 
grasslands soils. 
X-ray computed 
tomography and 
image analytics for 
soil columns 
Comprehensive 
measurement of 
macropore networks in 
soils from forests, with 
strong 
interconnectedness 
and vertical 
consistency 
Particular to forest 
soils; not directly 
tackling grassland 
landscapes. 
[14] To recreate scaffolds 
from a 
Data-driven 
approach involving 
Over 70% accuracy in 
recreating scaffolding 
Concentrated on 
construction areas, 

3D Reconstruction of Grassland Landforms Using Intelligent Robot…  Informatica 48 (2024) 179-190   181 
 
photogrammetric 
point cloud of 
building sites, 
utilizing a unique 
3D local feature 
description. 
preprocessing, 
classification, and 
geometric modeling 
with the LSSHOT 
descriptor. 
elements; excellent 
noise management 
may not directly 
applicable to 
natural landscape 
reconstruction. 
[15] Create an 
interpolation 
algorithm for 
dynamic 3D face 
rebuilding. 
Automated iterative 
interpolation 
algorithm for point 
cloud data. 
Excellent saturability 
and thoroughness in 
3D facial feature 
rebuilding. 
Particular to facial 
recognition and not 
immediately 
associated with 
geological surveys. 
[16] To present a 3D 
machine vision-
enabled smart robot 
architecture 
Automated 
calibration of 
monocular and 
binocular cameras, 
3D placement 
utilizing binocular 
vision. 
Efficient binocular 3D 
location in simple 
situations and a stable 
system 
Restricted to 
controlled 
surroundings, may 
struggle with 
complicated natural 
landscape. 
[17] To create a scheme 
for natural 
grassland-type 
recognition utilizing 
deep learning. 
PyTorch deep 
learning system 
with transfer 
learning from 
VGG-19 and data 
augmentation 
Optimum grassland 
identification model 
with excellent 
accuracy 
Depending on the 
quality of images 
and field data, it 
might not include 
all grassland types. 
[18] To create a smart 
grazing solution for 
unmanned animal 
husbandry. 
Incorporation of 
remote sensing, 
Internet of Things, 
and machine 
learning 
Effective grazing 
approach for better 
utilization of resources 
Concentrated on 
animal husbandry, 
not directly 
tackling geological 
landscape 
rebuilding. 
[19] To develop a QR 
code-based interior 
navigation system 
for attendant robots. 
QR code 
geolocation and 
planning of paths 
algorithms. 
Efficient indoor 
navigation for robots, 
with great accuracy 
Limited to interior 
surroundings; not 
suited for large-
scale outside 
surveys. 
[20] To enhance object 
detection in robot 
vision, 
discriminative bit 
selection hashing 
RGB-D object 
detection using the 
bit selection 
hashing approach 
Enhanced object 
recognition accuracy 
in robotic vision 
systems. 
Particular to object 
identification, does 
not deal with large-
scale landform 
analytics. 
The evaluated works show a wide range of methodology 
and uses in ecological tracking and landform assessment, 
including soil property assessment, sophisticated image 
evaluation, and 3D reconstruction methods. While each 
study provides useful insights, constraints such as a focus 
on particular areas, settings, or architecture highlight the 
need for more flexible and broad techniques. The 
suggested work uses a smart robot vision system to tackle 
these constraints, giving a better solution for precise and 
efficient 3D reconstruction of grassland landscapes. 
 
2.1 Three-dimensional reconstruction 
method of grassland landform 
Grassland landform is a kind of ecological environment. 
The reason for the formation of grassland landform is that 
the soil layer is thin and the precipitation is less, so that 
most herbs cannot grow. The structure of grassland 
landforms is complex and  

182    Informatica 48 (2024) 179-190   M. Zhong et al. 
 
Steppe landform
Three-dimensional 
reconstruction
grassland structure
Geomorphic features
Safeguard
 
 
Figure 1: Process map of 3D reconstruction of grassland 
landforms 
 
changeable. By 3D reconstruction of grassland landforms, 
the data characteristics of grassland landforms can be 
obtained, so as to protect grassland resources in a targeted 
manner. The 3D reconstruction process of the grassland 
landform is shown in Figure 1. 
The three-dimensional image analysis approach 
necessitates a simulation to validate the suggested 
rationality of the three-dimensional vision-based 
distribution of grassland landform. A simulation platform 
for three-dimensional image analysis is developed under 
the Windows 10 environment to ensure that landscape 
gardens and grassland landform are distributed rationally. 
The Point Grey Flea2 camera was utilized, and the 
experimental data set consisted of fifty frames of photos 
supplied by the Association of Landscape Architecture. 
The resolution of the camera was 640pi × 520pi. The 
experimental PC host is equipped with the following 
hardware: a 3.0GHz Intel Core i5 8500 quad-core 
processor, 8 GB of RAM, a 4 GB Nvidia GTX970 
graphics card, and a 500 GB hard drive. Adobe 
Photoshop CS6, Auto CAD for 3D drawings, and 
SketchUp for 3D models are some of the platforms 
utilized. 
In Figure 1, the 3D reconstruction process of the 
grassland landform is as follows: photographing the 
grassland landform with a camera, extracting feature 
points, and using a feature matching algorithm to 
construct the grassland landform in 3D. Finally, 
according to the constructed three-dimensional model, 
the structure and characteristics of grassland landforms 
can be analyzed, and corresponding grassland protection 
measures can be formulated according to the landform 
characteristics. 
Extraction of 3D Features: The first step in 3D 
reconstruction of grassland landforms is to extract 3D 
features of grassland landforms, which is based on 
camera vision by analysing information that is salient and 
less likely to be lost in the grassland images captured by 
the camera. The 3D features of grasslands are mainly 
divided into three types: point, line, and flat. 
The matching of three-dimensional features is the 
premise of feature extraction. At present, the common 
three-dimensional feature matching is mainly through 
grayscale matching, and by establishing a window model. 
The feature matching search is performed at the center of 
the image. Before feature extraction, it is necessary to 
determine the position of the camera, and try to ensure 
that the brightness of each pixel of the grassland landform 
image is related to the intensity of the reflected light on 
its surface. There are two types of camera shooting: fixed 
camera shooting and motion camera shooting. 
Stereoscopic vision of grassland landforms can be 
obtained through different grassland angles captured by 
multiple cameras, and three-dimensional information of 
grassland landforms can be obtained through the 
deviation of image pixel positions under the vision of 
different cameras. The principle of grassland stereo 
vision is shown in Figure 2. 
A
A
0
A
1
A
2
B
1
B
2
D
1
D
2
 
Figure 2: Schematic diagram of grassland stereo vision 
In Figure 2, assuming that there are two grassland 
landform photographing cameras 
1
B and 
2
B , and A is 
any point in the grassland landform space, 
1
A and 
2
A 
are the imaging points of point A in the two cameras 
respectively. Observing point, A from the perspectives of 
two different cameras, there can be two paths A B
1
 and 
A B
2
 , but there is a common point A under the 
perspectives of the two cameras, so point A is a 3D point 
under the camera perspective. 
The principle of 3D stereo measurement is as follows. 
Assuming that the projection matrices of cameras 
1
B 
and 
2
B  are 
1
R  and 
2
R  respectively, and the 
coordinates of point A in the coordinate systems of the 
two camera perspectives are ) , (
1 1 1
b a A  and 
) , (
2 2 2
b a A respectively, the following relationship can 
be obtained: 

3D Reconstruction of Grassland Landforms Using Intelligent Robot…  Informatica 48 (2024) 179-190   183 
 






















=










1
   
   
   
1
34
1
33
1
32
1
31
1
24
1
23
1
22
1
21
1
14
1
13
1
12
1
11
1
1
1
1
Z
Y
X
r r r r
r r r r
r r r r
b
a
ZD         (1) 
In Formula (1), 
1
D is the abscissa in the camera 
1
B . 
Similarly, the coordinate relationship under camera 
2
B 
can be obtained. 






















=










1
   
   
   
1
34
2
33
2
32
2
31
2
24
2
23
2
22
2
21
2
14
2
13
2
12
2
11
2
2
2
2
Z
Y
X
r r r r
r r r r
r r r r
b
a
ZD     (2) 
Combining Formula (1) and Formula (2), Formula (3) can 
be obtained: 
( 𝑎 1
𝑟 1
31
− 𝑟 1
11
) 𝑋 + ( 𝑎 1
𝑟 1
32
− 𝑟 1
12
) 𝑌 + ( 𝑎 1
𝑟 1
33
− 𝑟 1
13
) 𝑍 =
𝑟 1
14
− 𝑎 1
𝑟 1
34
                        (3) 
Through Formula (3), the values of X, Y , and Z can be 
obtained, so that the three-dimensional coordinates of any 
point in the grassland landform can be obtained. 
Point feature extraction algorithm 
The image features of grassland landforms are data that 
can effectively reflect the features of grassland landforms. 
By analyzing the characteristic data, the calculation 
amount of 3D reconstruction can be reduced, and a lot of 
non-feature memory can be saved. The traditional 3D 
feature extraction method is point feature extraction. 
By calculating the gradient of each pixel in the grassland 
landform image, and calculating the point with the 
maximum value of the two gradient directions, the 
detection process of point feature extraction is as follows: 
2
)] ( [ ) det( P ur k P Q − =       (4) 
In Formula (4), when the value of Q exceeds the set 
interval, Q is the feature point. P represents a pixel in the 
grassland landscape. 
Gradient direction
level one
Second floor
the third floor
fourth floor
fifth floor
 
Figure 3: Process diagram of point feature extraction 
 
The process of point feature extraction is shown in Figure 
3. 
In Figure 3, by continuously extracting pixel gradient 
feature points, feature points that can summarize the 
measured image can be extracted after multiple feature 
extractions, and finally the grassland landform can be 
reconstructed with the feature points. 
 
2.2 Intelligent robot vision technology 
The intelligent robot vision system combines visual 
analysis, three-dimensional construction with robots, 
with the functions of autonomous positioning and 
intelligent analysis, which can be effectively applied to 
3D reconstruction of grassland landforms. 
In the three-dimensional reconstruction of the intelligent 
robot vision system, the image problem in motion needs 
to be considered. During the movement of the robot, the 
camera worn by the robot can rotate and shake to different 
degrees. It is necessary to effectively analyze the feature 
data during 3D reconstruction. The structure of the 
feature extraction  

184    Informatica 48 (2024) 179-190   M. Zhong et al. 
 
Feature points
Block
Edge
 
 
Figure 4: Structure diagram of feature extraction 
 
of the intelligent robot vision system is shown in Figure 
4. 
In Figure 4, the structure of feature extraction is mainly 
divided into feature points, blocks and edges. Uncertainty 
analysis of image rotation is mainly implemented in the 
form of determining feature points. The selection of 
feature points is mainly through the analysis of the 
centroid point of the image. 
The moment of A in a grassland image A is supposed to 
be expressed as Formula (5): 


=
A y x
b a
b a
y x r
,
,
         (5) 
In Formula (5), the value range of a and b is } 1 , 0 { . 
Using the moments of image A, the centroid of image A 
can be found: 
) , (
00
01
00
10
r
r
r
r
S =         (6) 
In Formula (6), S represents the centroid of the image A. 
Supposing that the geometric center of image A be point 
G, then the feature vector is GS , and the direction of 
the feature point can be obtained. 
) tan(
10
01
r
r
acr =          (7) 
In the process of feature point extraction, the intelligent 
robot vision system can determine the angle of the feature 
point, thereby determining the position of the feature 
point, and improving the stability and accuracy of 
processing grassland landform images. 
 
3  Three  dimensional 
reconstruction experiment and result 
 
3.1 Three-dimensional reconstruction 
experiment of grassland landform 
 
3D Reconstruction Data of Grassland Landforms: Three-
dimensional reconstruction of grassland landforms is a 
data analysis of grassland landforms, which is conducive 
to accurately studying the structural composition of 
grasslands and the stability of grassland ecosystems. To 
carry out 3D reconstruction analysis of grassland 
landforms, it is necessary to establish an index that can 
be used to evaluate the effect of 3D reconstruction of 
grassland landforms. For this purpose, the experiment 
conducted statistics on the close contacts in the 3D 
reconstruction system of grassland landforms [21]. Table 
2 shows the personnel involved in the analysis of 
grassland landforms. 
 
Table 2: Information about the personnel involved in the 
analysis of grassland landforms 
 
Serial 
number 
Person type 
Number 
of people 
(person) 
Proportion 
1 
Prairie 
landscape 
surveyor 
80 32% 
2 
3D 
reconstruction 
technician 
90 36% 
3 
Grassland 
landform 
photographer 
80 32% 
 
In Table 2, the survey data of relevant personnel in the 3D 
reconstruction of grassland landforms are described. The 
survey objects were divided into three categories, namely 
grassland landform surveyors, 3D reconstruction 
technicians, and grassland landform photographers. The 
largest number of people were 3D reconstruction 
technicians, with the proportion of 36%. 
A questionnaire survey was conducted on the above-
mentioned close contacts of the grassland landforms, and 
the evaluation indicators of the three-dimensional 
reconstruction of the grassland landforms were mainly 
counted, as well as the proportion of the number of people 
each index accounted for [22]. The statistical results of 
the evaluation indicators for the above 250 people are 
shown in Table 3. 
 
Table 3: Statistical results of evaluation indicators 
 
Serial Evaluation Number Proportion 

3D Reconstruction of Grassland Landforms Using Intelligent Robot…  Informatica 48 (2024) 179-190   185 
 
number indicators of people 
(person) 
1 
Accuracy of 3D 
Reconstruction 
65 26% 
2 
Feature data 
matching 
degree 
60 24% 
3 
Noise 
Immunity for 
3D 
Reconstruction 
55 22% 
4 
3D 
reconstruction 
time 
60 24% 
5 
Memory 
required for 3D 
reconstruction 
10 4% 
 
In Table 2, a total of 5 evaluation indicators were counted, 
of which the accuracy indicator of 3D reconstruction had 
the largest proportion, with the proportion of 26%. The 
smallest proportion was the memory index required for 
3D reconstruction, which accounted for 4%. Due to the 
large amount of memory that can be endured when 
reconstructing grassland landforms, the memory space 
required for 3D reconstruction had little effect on 3D 
reconstruction. Therefore, the first four indicators in 
Table 2 were used as indicators for evaluating the effect 
of 3D reconstruction of grassland landforms [23]. 
Design Process of 3D Reconstruction of Grassland 
Landforms: In order to analyze the effect of 3D 
reconstruction of grassland landforms based on the 
intelligent robot vision system, a comparison and analysis 
was made with the traditional 3D reconstruction of 
grassland landforms. Among them, the traditional 3D 
reconstruction adopted point feature extraction algorithm 
for analysis. In order to ensure the accuracy of the two 3D 
reconstruction comparison experiments, the experimental 
environment of the two 3D reconstruction methods must 
be the same, and the same grassland landform must be 
analyzed. 
In order to ensure the vastness and diversity of grassland 
landforms, the experiment took the Qinghai-Tibet 
grassland area in China as an example. When comparing 
the two 3D reconstruction methods, in order to ensure the 
integrity of the comparison experiment, the experiment 
set up grassland areas of different sizes for analysis. 
Experiments were performed on a square area with a side 
of 2km and a square area with a side of 4km [24]. 
In the process of 3D reconstruction of grassland 
landforms, the data of one experiment was serendipitous. 
To this end, the experiment carried out the three-
dimensional reconstruction process for several times for 
the comparison test of the two three-dimensional 
reconstructions, which was set to 1st, 2nd, 3rd, 4th, 5th 
and 6th, and observed the effects of the two three-
dimensional reconstructions under each number of times. 
 
3.2 Results of 3D reconstruction comparison 
experiments 
Accuracy of 3D Reconstruction: The effect of 3D 
reconstruction is mainly determined by the accuracy of 
3D reconstruction. The higher the accuracy of 3D 
reconstruction, the better the effect of 3D reconstruction 
of grassland landforms. The basis for judging the 
accuracy of 3D reconstruction is the similarity between 
the grassland landform after 3D modeling and the actual 
grassland landform. The 3D reconstruction accuracy of 
the 3D reconstruction of the two grassland landforms was 
compared, in which six comparison results were set. The 
comparison results of the accuracy of the 3D 
reconstruction of the two grassland landforms are shown 
in Figure 5. 
 
 
(a) Square grassland with a side of 2km          
 
 
0%
20%
40%
60%
80%
100%
120%
1 2 3 4 5 6
Accuracy
Number of comparisons
Traditional mode
Intelligent robot mode
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1 2 3 4 5 6
Accuracy
Number of comparisons
Traditional mode
Intelligent robot mode

186    Informatica 48 (2024) 179-190   M. Zhong et al. 
 
(b) Square grassland with a side of 4km 
 
Figure 5: 3D reconstruction accuracy comparison results 
 
In Figure 5(a), a comparison of two 3D reconstruction 
methods is depicted on a square grassland with a side of 
2 km. In the traditional 3D reconstruction method, the 
accuracy of 3D reconstruction first increased and then 
decreased during the six experiments, while the accuracy 
of grassland 3D reconstruction based on intelligent robot 
vision system generally showed an upward trend. In the 
sixth experiment, the accuracies of the two 3D 
reconstructions were 72% and 96% respectively. In 
Figure 5(b), the 3D reconstruction accuracy experiment 
was performed on a square grassland area with a side of 
4km. The accuracy of the traditional 3D reconstruction of 
grassland landforms changed significantly, which started 
to drop after the second experiment, while the accuracy 
of the 3D reconstruction based on the intelligent robot 
vision system did not change much. In the sixth 
experiment, the accuracies of the two 3D reconstructions 
were 52% and 89% respectively. Therefore, the 3D 
reconstruction of grassland based on intelligent robot 
vision system can effectively improve the accuracy of 3D 
reconstruction. 
Matching Degree of Feature Data: The three-dimensional 
reconstruction of grassland landforms is mainly through 
matching and analysis of feature data, and the feature data 
matching of traditional three-dimensional reconstruction 
methods adopts point feature extraction and matching 
technology. The 3D reconstruction of the intelligent robot 
vision system uses intelligent robot vision technology for 
feature matching. The matching degree of feature data 
was compared between the two 3D reconstruction 
methods, and was compared for 6 times in the experiment. 
The comparison result of the matching degree of feature 
data is shown in Figure 6. 
 
 
(a) Square grassland with 2km side 
 
 
(b) Square grassland with 4km side 
 
Figure 6: Comparison result of feature data matching 
degree 
 
In Figure 6(a), it was a comparison of characteristic data 
of three-dimensional reconstruction in a square grassland 
area with a side of 2 km. The experiments were carried 
out 6 times, and the matching degree of the characteristic 
data of the traditional 3D reconstruction of grassland 
landforms was lower than that of the intelligent robot 
vision system. In the six comparison results, the average 
matching degrees of the two 3D reconstruction feature 
data were 73.3% and 89.3% respectively. In Figure 6(b), 
the grassland landform compared in the experiment was 
a square grassland with a side of 4km. The difference in 
the characteristic data of the two 3D reconstruction 
methods was more obvious. The average matching 
degree of the characteristic data of the two 3D 
reconstruction methods was 53.3% and 85.7% 
respectively. Therefore, applying the intelligent robot 
vision system to the 3D reconstruction of grassland 
landforms can improve the matching degree of feature 
data [25]. 
Noise Immunity for 3D Reconstruction: During 3D 
reconstruction of grassland landforms, it can be affected 
by many factors, including camera shooting errors, 
feature data positioning errors, etc., which affect the 
effect of 3D reconstruction. The noise resistance of 3D 
reconstruction is the embodiment of the stability of 3D 
reconstruction. The two 3D reconstruction methods were 
used for 6 experiments on two different grassland areas, 
and the noise immunity of two types of 3D 
reconstructions was compared. The comparison results of 
noise immunity of 3D reconstruction are shown in Figure 
7. 
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1 2 3 4 5 6
Matching degree
Number of comparisons
Traditional mode
Intelligent robot mode
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1 2 3 4 5 6
Matching degree
Number of comparisons
Traditional mode
Intelligent robot mode

3D Reconstruction of Grassland Landforms Using Intelligent Robot…  Informatica 48 (2024) 179-190   187 
 
 
 
Figure 7: Comparison results of 3D reconstruction noise 
immunity 
 
In Figure 7, the overall data fluctuation of the three-
dimensional reconstruction of the two grassland 
landforms was not large under the results of the six 
experiments. However, the six experimental data results 
were based on the intelligent robot vision system, and the 
noise immunity of 3D reconstruction was greater than 
that of traditional 3D reconstruction. Among them, in the 
fourth experiment, the difference in noise immunity 
between the two was the largest, and the traditional three-
dimensional noise immunity at this time was 68%, while 
the noise immunity of 3D reconstruction based on 
intelligent robot vision system was 92%. In the first set of 
experiments, the noise immunity of the two 3D 
reconstruction methods was the smallest, with a 
difference of 10%. Therefore, the intelligent robot vision 
system can improve the noise resistance of 3D 
reconstruction of grassland landforms. 
Time of 3D Reconstruction: The time required for 3D 
reconstruction refers to the time required to complete the 
entire process from camera shooting and framing to 
feature data matching to 3D modeling [26-27]. The time 
of 3D reconstruction of grassland landform reflects the 
efficiency of 3D reconstruction. The shorter the time of 
3D reconstruction, the higher the  
 
(a) Square grassland with 2km side              
 
(b) Square grassland with 4km side 
 
Figure 8: Comparison of 3D reconstruction accuracy 
results 
 
efficiency of 3D modeling. The time required for two 3D 
reconstructions of grassland landforms was compared, 
and the comparison results of 3D modeling 
reconstruction time are shown in Figure 8. 
In Figure 8(a), it shows the time required for two 3D 
reconstruction methods to perform 3D reconstruction on 
a square grassland with a side of 2km. Among them, the 
traditional 3D reconstruction time was 3.2 hours at the 
lowest, 4.0 hours at the highest, and 3.8 hours on average 
during the six tests, while the minimum time required for 
3D reconstruction based on the intelligent robot vision 
system was 1.6 hours and the maximum was 2.6 hours, 
with the average of 2.0 hours. In Figure 8(b), a square 
grassland with a side of 4km was used as the 
experimental area, and the time required for the 3D 
reconstruction based on the intelligent robot vision 
system was significantly shorter than that of the 
traditional 3D reconstruction. In the fifth test, the time 
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
1 2 3 4 5 6
Noise immunity
Number of comparisons
Traditional mode
Intelligent robot mode
0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
1 2 3 4 5 6
Reconstruction time(h)
Number of comparisons
Traditional mode
Intelligent robot mode
0
1
2
3
4
5
6
7
8
1 2 3 4 5 6
Reconstruction time(h)
Number of comparisons
Traditional mode
Intelligent robot mode

188    Informatica 48 (2024) 179-190   M. Zhong et al. 
 
required for the two 3D reconstructions differed the most, 
with a difference of 4.0 hours. Therefore, the 3D 
reconstruction of grassland landforms based on the 
intelligent robot vision system can greatly shorten the 
time of 3D reconstruction and improve the efficiency of 
3D reconstruction of grassland landforms. 
Taking everything into account, the system has shown to 
be quite stable, which is crucial for ensuring accurate and 
precise image recognition. However, it is clear that the 
loading speed is affecting the system's access speed. 
Given the wide range of image recognition functions, 
there is no one-size-fits-all solution when it comes to 
meeting the function's individual criteria. Ultimately, it is 
necessary to integrate and present the data from all 
functions together, and progress will be hindered by 
specific slower processing functions. Alternatively, when 
it comes to server indicators, specific cluster indices 
could vary depending on server characteristics and 
regional network conditions. These discrepancies are 
now within a more manageable range, but there is room 
for improvement. As a whole, there is room for 
improvement in all of this system's modules. 
 
4  Conclusion 
The three-dimensional reconstruction of grassland has a 
positive effect on exploring the landform structure of 
grassland and promoting the development of grassland. 
In this paper, the intelligent robot vision system was 
applied to the 3D reconstruction of grassland landforms, 
and the efficiency of 3D reconstruction was improved by 
optimizing the matching of grassland landform feature 
data. By comparing with the traditional 3D reconstruction 
of grassland landforms, the experimental results showed 
that the application of the intelligent robot vision system 
to the 3D reconstruction of grassland landforms can 
effectively improve the accuracy of 3D reconstruction, 
and improve the matching degree of feature data and the 
stability of 3D reconstruction. The efficiency of 3D 
reconstruction can be improved by the matching and 
optimization of characteristic data through intelligent 
technology, which can not only conduct more detailed 
analysis of the grassland landform structure, but also 
predict the development trend of the grassland ecological 
environment, as well as protect and develop the grassland 
ecological environment in a targeted manner. However, 
the comparison of the two 3D reconstructions in this 
paper was only limited to the square grassland area with 
side lengths of 2 km and 4 km, and no comparison was 
made for larger or smaller grassland areas. Therefore, 
expanding the grassland area for the comparison of the 
two 3D modeling methods is the direction of future 
research. 
 
Data availability 
The data used to support the findings of this study are 
available from the corresponding author upon request. 
 
Conflicts of interest 
The authors declare no conflicts of interest. 
 
Funding statement 
This research study is sponsored by Hebei Provincial 
Department of Education Youth Top Talent Project. The 
name of the project is Characteristics of near-surface 
wind field of grassland landform and wind resistance 
performance of photovoltaic panels. The project number 
is BJ202010. Thank the project for supporting this article! 
 
References 
[1] Ko J H, 2020 “Robot Vision System based on 3D 
Depth map and Object Recognition”, Journal of the 
Institute of Electronics and Information Engineers, 
vol.57, no.3, pp.101-105.. 
[2] Park D ,  Kang K M ,  Bae J W , Han J H, 2019. 
“Robot Vision to Audio Description Based on Deep 
Learning for Effective Human-Robot Interaction”, 
The Journal of Korea Robotics Society, vol.14, no.1, 
pp.22-30. 
[3] Kim J M ,  Choi C W ,  Jang W S, 2018. 
“Development of Robot Vision Control Schemes 
based on Batch Method for Tracking of Moving 
Rigid Body Target”, Journal of the Korean Society of 
Manufacturing Process Engineers, vol.17, no.5, 
pp.161-172. 
[4] Cao T Y ,  Cai H Y ,  Fang D M ,  Liu C, 2017. 
“Robot vision system for keyframe global map 
establishment and robot localization based on 
graphic content matching”, Guangxue Jingmi 
Gongcheng/Optics and Precision Engineering, 
vol.25, no.8, pp.2221-2232. 
[5] Wang J ,  Tang J ,  Guo T ,  Zhang S ,  Xia W ,  
Tan H , et al, 2019. “C3N4-digested 3D construction 
of hierarchical metallic phase MoS2 nanostructures”, 
Journal of Materials Chemistry A, vol.7, no.31, 
pp.18388-18396. 
[6] Chen C ,  Tang L ,  Hancock C M ,  Zhang P, 
2019. “Development of low-cost mobile laser 
scanning for 3D construction indoor mapping by 
using inertial measurement unit, ultra-wide band and 
2D laser scanner”, Engineering Construction & 
Architectural Management, vol.26, no.7, pp.1367-
1386. 
[7] Roslan A F ,  Salehuddin F ,  Zain A ,  Kaharudin 
K E ,  Maheran A, 2018. “30nm DG-FinFET 3D 
construction impact towards short channel effects”, 
Indonesian Journal of Electrical Engineering and 
Computer Science, vol.12, no.3, pp.1358-1365. 
[8] Asa K ,  Funabora Y ,  Doki S ,  Doki K, 2017. 
“Automatic Measuring Position Determination of 

3D Reconstruction of Grassland Landforms Using Intelligent Robot…  Informatica 48 (2024) 179-190   189 
 
UA Vs for Visual Inspection System of 
Infrastructures Based on 3D Construction Model 
Considering Accuracy of Data”, Transactions of the 
Society of Instrument & Control Engineers, vol.53, 
no.3, pp.229-235. 
[9] Mukundan R, 2017. “Contour Based High 
Resolution 3D Mesh Construction Using HRCT and 
MRI Stacks”, International journal of multimedia 
data engineering & management, vol.8, no.4, pp.60-
73. 
[10] Patil M C ,  Patil S S, 2020. “3D Construction of 
Brain using 2D MRI Slices”, Gender Technology and 
Development, vol.9, no.4, pp.2278-3075. 
[11] Huang Y ,  Kuang X ,  Cao Y ,  Bai Z, 2018. “The 
soil chemical properties of reclaimed land in an arid 
grassland dump in an opencast mining area in China”, 
RSC Advances, vol.8, no.72, pp.41499-41508. 
[12] Sun R ,  Xu W D ,  Chen W ,  Zhang Y , 2017. 
“The quantitative characteristics of plant germplasm 
resources in Horqin grassland”, Chinese Journal of 
Ecology, vol.36, no.10, pp.2699-2706. 
[13] Zhang J M ,  Xu Z M ,  Li F ,  Hou R J ,  Ren Z, 
2017. “Quantification of 3D macropore networks in 
forest soils in Touzhai valley(Yunnan, China) using 
X-ray computed tomography and image analysis”, 
Journal of Mountain Science, vol.14, no.3, pp.474-
491. 
[14] Xu Y ,  Tuttas S ,  Hoegner L ,  Stilla U, 2018. 
“Reconstruction of scaffolds from a 
photogrammetric point cloud of construction sites 
using a novel 3D local feature descriptor”, 
Automation in construction, vol.85, no.1, pp.76-95. 
[15] Sui D ,  Hou D ,  Duan X, 2017. “An interpolation 
algorithm fitted for dynamic 3D face reconstruction”, 
Multimedia Tools and Applications, vol.76, no.19, 
pp.19575-19589. 
[16] Zhang Y ,  Zhao J ,  Han H, 2021. “A 3D Machine 
Vision-Enabled Intelligent Robot Architecture”, 
Mobile Information Systems, vol.2021, no.3, pp.1-11. 
[17] Cheng C ,  Guo X ,  Chen X ,  Li C ,  Xiong H, 
“Control System Development of Intelligent Robot 
Based on Mindstorms”, Wuhan Ligong Daxue 
Xuebao (Jiaotong Kexue Yu Gongcheng 
Ban)/Journal of Wuhan University of Technology 
(Transportation Science and Engineering), vol.42, 
no.2, pp.247-252, 2018. 
[18] Park S, 2018. “Computational Vision for Automatic 
Tracking and Objective Estimation of Mobile Robot 
Trajectory”, Journal of Computer Sciences and 
Applications, vol.6, no.1, pp.17-22. 
[19] Wang J ,  Xiao X, 2018. “Mobile robot integrated 
navigation method based on QR code vision 
positioning”, Chinese Journal of Scientific 
Instrument, vol.39, no.8, pp.230-238. 
[20] Feng L ,  Liu Y ,  Li Z ,  Zhang M ,  Wang F ,  
Liu S , et al, 2019. “Discriminative bit selection 
hashing in RGB-D based object recognition for robot 
vision”, Assembly Automation, vol.39, no.1, pp.17-
25. 
[21] Borges F ,  Fernandes R ,  Silva I N ,  Silva C B 
S, 2017. “Feature Extraction and Power Quality 
Disturbances Classification Using Smart Meters”, 
IEEE Transactions on Industrial Informatics, vol.12, 
no.2, pp.824-833. 
[22] You M ,  Wang H ,  Liu Z ,  Chen C ,  Liu J ,  
Xu X H , et al, 2017. “Novel feature extraction 
method for cough detection using NMF”, Iet Signal 
Processing, vol.11, no.5, pp.515-520. 
[23] Mo D ,  Lai Z ,  Wong W, 2019. “Locally Joint 
Sparse Marginal Embedding for Feature Extraction”, 
IEEE Transactions on Multimedia, vol.21, no.12, 
pp.3038-3052. 
[24] Hang R ,  Liu Q ,  Sun Y ,  Yuan X ,  Pei H ,  
Plaza A , et al, 2017. “Robust Matrix Discriminative 
Analysis for Feature Extraction From Hyperspectral 
Images”, IEEE Journal of Selected Topics in Applied 
Earth Observations and Remote Sensing, vol.10, 
no.5, pp.2002-2011. 
[25] Jung J Y ,  Kim S W ,  Yoo C H ,  Park W J ,  Ko 
S J, 2017. “LBP-ferns-based feature extraction for 
robust facial recognition”, IEEE Transactions on 
Consumer Electronics, vol.62, no.4, pp.446-453. 
[26] Jurczyk-Bunkowska M, 2021. “Tactical 
manufacturing capacity planning based on discrete 
event simulation and throughput accounting: A case 
study of medium sized production enterprise”, 
Advances in Production Engineering & Management, 
vol. 16, no. 3, pp.335-347. 
[27] Zhang Y D ,  Liao L ,  Yu Q ,  Ma W G ,  Li K 
H, 2021. “Using the gradient boosting decision tree 
(GBDT) algorithm for a train delay prediction model 
considering the delay propagation feature”, 
Advances in Production Engineering & Management, 
vol. 16, no. 3, pp.285-296. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

190    Informatica 48 (2024) 179-190   M. Zhong et al.