2020 International Conference on Electrical Engineering (ICEE) September 25-27, 2020, Istanbul, Turkey

UAV Path Planning Based on Dynamic

Programming Algorithm On Photogrammetric
DEMs
Adel MOKRANE, Amal CHOUKCHOU BRAHAM, Brahim CHERKI
Laboratoire d’Automatique de Tlemcen, LAT
University of Tlemcen
Tlemcen, Algeria
[email protected], [email protected], [email protected]

Abstract—The aim of this paper is to propose a three- like stereo-cameras, laser rangers, range cameras, ultrasonic
dimensional path planning algorithm based on dynamic pro- rangers . . . where the aim is build 2D or 3D models of the
gramming to plan optimal trajectories for an Unmanned Aerial environment [2]. One of the most important models is the
Vehicle (UAV) on a Digital Elevation Model (DEM) map. An RGB
camera is employed to obtain point cloud data of the map using Digital Elevation Map or DEM. This latter can be created
WebODEM. The point cloud data are processed in to a digital using laser scanners. In our approach, a digital camera was
elevation map using GRASS-GIS. Then, dynamic programming used to create DEMs that can be identical to DEMs created
approach is used to find an optimal path from a starting location by laser scanners. This can be done using photogrammetry.
to a target location. The effectiveness and practicality of this
paper work are demonstrated by the simulation test results.
In the recent years, UAVs have been utilized in the field of
Index Terms—path planning, dynamic programming, optimal geomatics. The UAV photogrammetry has introduced a low-
path, digital elevation model, UAV. cost alternative to the classical manned aerial photogrammetry
and opened several and new approached in the close-range
aerial domain [3]. UAV photogrammetry is a photogrammet-
I. I NTRODUCTION ric measurement platform that can be autonomously, semi-
autonomously or remotely controlled. The platform can be
UAVs, which stand for Unmanned Aerial Vehicle or better provided with photogrammetric measurement equipment like
known as drones, are flying machines that can perform au- video or digital camera, thermal (IR) camera, airborne Li-
tonomously repetitive tasks in stressful environments with high DAR (Light Detection and Ranging) system or a combination
precision and accuracy. UAVs are robust devices that come of thereof [4]. UAV Photogrammetry has been an essential
with low cost and weight. They can be used for better decision means in the cartographic base production all over the world.
making and are able to fly into zones which would be dan- Such technique has allowed the achievement of innovative
gerous for pilots. Thanks to the aforementioned advantages, results using advanced techniques for treating digital images
UAVs are gaining more in popularity and interest. Therefore, [5]. The current processors, which are developed with high-
they have been employed in many domestic applications that performance, have aided to the enhancement and execution of
vary from structural inspection, photography, surveillance, the instruments used in image processing of the photogram-
environmental explorations to precision agriculture . . . etc. metry systems.

In many applications, path planning is an important element In this study, UAV path planning based on DEM map is
of the UAV systems when defining their missions. UAVs presented. In section II, the DEM based on UAV photogram-
should intelligently plan their own paths for avoiding the metry will be introduced. In section III, robot path planning
obstacles present in the environment where their missions will be discussed, in general. Section IV will be dedicated for
are assigned. Generally speaking, path planning consists of explaining the planner used in this work. The remainder will
identifying a path that will cause the UAV to reach the be devoted for the experimental simulations and their results.
goal location and to avoid the collision with obstacles in the
environment when executed [1]. However before the path is
planned, the UAV has to have complete or partial information II. ROBOT PATH P LANNING
about the environment where it will navigate. Therefore, map- RPP related problems have been widely studied by many
ping the environment is a prior step before planning the path. researchers, generally focusing on manipulators and ground
The mapping process can be performed by several sensors robots. From a technical point of view, path planning is a

978-1-7281-8292-6/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: London School of Economics & Political Science. Downloaded on May 17,2021 at 14:21:43 UTC from IEEE Xplore. Restrictions apply.
 2020 International Conference on Electrical Engineering (ICEE) September 25-27, 2020, Istanbul, Turkey

problem of finding a path for an agent in a properly defined weighted multistage graph in backward fashion, and sorts the
environment from a starting point to a goal point such that the minimum path in forward fashion. Hence, the forward dynamic
agent is free from collisions with surrounding obstacles and programming approach has a backward reasoning. On the
its planned trajectory satisfies the physical and the kinematic other hand, the backward approach computes the minimum
constraints of that agent [6]. RPP can be global or local. cost function in forward fashion and sorts the minimum path
Global RPP is performed prior to the agent movement. It in backward fashion. The backward dynamic programming
utilizes the information from the agent’s surrounding word method has a forward reasoning.
to reach the goal location from the starting one. As the
information includes global data, hence the process is slow, Pseudocode Forward Dynamic Programming
however the path generated may be optimal. Local RPP refers 1. Algorithm FDP(G,E,k,n,P)
to the process of avoiding obstacles by the means of the 2. The input is a k stage graph G with n vertices
acquired information from the current robot’s surrounding 3. E is set of edges and c[i,j] is the cost of <i,j>
environment. RPP takes into account some important criteria 4. P[1:k] is the minimum cost
such as path length, computational time and completeness [7]. 5. {
In our case, the generated path must be optimal, hence this 6. cost[n]=0;
will minimize the UAV flight time. In other words, it prolongs 7. for j = n-1 to 1
the UAV’s endurance, minimizes the energy consumption, and 8. {
decreases revelation to possible risks. In literature, there are 9. for l = j+1 to n
many methods to find for solving the UAV path planning 10. {
problem. An UAV system in [8] used an improved artificial 11. /where l is a vertex such that <i,l>is
potential field algorithm for generating a path while tracking 12. an edge of G
a dynamic target. In addition, the proposed method solved 13. and c[j,l]+cost[l] is minimum;
the local minimum problem. The study in [9] proposed a 14. cost[j]=c[j,l]+cost[l];
new path planning algorithm for reducing paths computing 15. d[j]=l;
time without path lengths degrading using visibility graph in 16. }
3D environment models with fixed convex obstacles. Better 17. }
qualities of paths were generated by using the obstacle ge- 18. //obtain the minimum cost path
ometry to explore only the desired region and avoid unwanted 19. P[1]=1; P[k]=n;
regions. Randomly sampling search algorithms were also used 20. for j = 2 to k-1 do
such as in [10] where Rapidly-exploring Random Trees RRTs 21. P[j]=d[P[j-1]];
were used to generate real time non-conflicting paths for 22. }
multiple UAVs in presence of static and dynamic obstacles.
On the other hand, graph search algorithms were widely used In our case, the forward dynamic programming approach can
for solving the path generation problem for UAV platforms. be used for solving the path planning of the UAV on the
Algorithms like Dijkstra [11] and A* [12] were the mostly raster data or DEM. DP computes the distance from all nodes
used by researchers. of the map to the goal. Firstly, it computes the distance to
nearer nodes (subproblem) known as pre-computed distances.
III. DYNAMIC P ROGRAMMING PATH P LANNING Only one step can be taken at a time to the surrounding
The technique of Dynamic Programming DP offers a nodes (up, down, right, left, upper right, upper left, lower
powerful method for solving problems that exhibit optimal right and lower left). As a result, a directed graph G = (V,
substructure. This latter is known as principle of optimality E) from the starting node to the target node is generated. The
where optimal solutions to related subproblems are contained directed graph contains vertices V which are partitioned into k
inside an optimal solution to the problem [13]. Dynamic stages and connected with edges E (distance between adjacent
programming is an optimization technique that transforms nodes). In addition, a transition cost matrix denoted by c(i,
complex problems into simple ones and forms stages where j), which contains the edge distance from one node in the
at each stage, a decision is made. The decisions at all the map to the rest of other nodes, is calculated over the raster.
stages form a set of interrelated and consecutive decisions Secondly, both the minimum cost function cost(i, j) and the
which guide to the final decision. Hence, such approach has a minimum distance d(j) are computed. Recurrence is used in
characteristic of multistage on optimization procedure. In other order to show the relationship between the problem and its
words, solving the global problem requires solving a large subproblems. Therefore using iteration i+1, the optimal path
number of subproblems. The unique solutions of these sub- is computed from any node by computing the optimal path of
problems are stored and may be fetched when needed. There neighboring nodes at i iteration. The optimal path of a node is
are two kinds of problem solving approaches: forward and updated every time. The minimum cost function is computed
backward dynamic programming [14]. The difference between using the following formula:
the two approaches is that the forward approach computes the
minimum cost function (or cost of the path) of the directed cost(i, j) = min{c(i, j) + cost(i + 1, l)} (1)

Authorized licensed use limited to: London School of Economics & Political Science. Downloaded on May 17,2021 at 14:21:43 UTC from IEEE Xplore. Restrictions apply.
 2020 International Conference on Electrical Engineering (ICEE) September 25-27, 2020, Istanbul, Turkey

where: i is the stage number, j is the different edges available

at a particular stage and l is the vertices away from the vertex
j. The minimum distance is simply taken as d(j) = l. Finally,
the minimum cost path p is sorted. The table below shows the
algorithm of the forward dynamic programming approach.

IV. S OFTWARE
To solve the UAV path planning problem, three software
were used. The first software one is WebODM which is
an open source, user-friendly and extendable application for
UAV image processing. It turns the captured images into
three dimensional geographic data. WebODM can be seen
as a toolchain for processing raw UAV imagery to other
useful products such as point clouds, Digital Surface Model
DSM, textured DSM, orthorectified imagery, ... etc. In our
previous work in [15], the use of WebODM was examined for
generating cost effective Digital Elevation Models from aerial
images. However in this case, WebODM is used for creating
a point cloud of the mapped region. The second software is Fig. 1. The methodology workflow.
GRASS GIS, which stands for Geographic Resources Anal-
ysis Support System. It is a Geographic Information System
GIS software. It is a free open-source software that can be
employed for managing and analyzing the geospatial data, D. Processing
processing images, producing graphics and maps, modeling This work is build of different procedure. The first thing
and visualizing spatial and temporal data. The third software to do is to plan an UAV flight to acquire images of the
is MatLAB R2016a which is used in order to generate the environment, where the UAV will navigate afterwards, using
optimal path from the starting node to the target node. an RGB camera attached to the UAV system. These RGB
images were uploaded to a computer for geo-tagging and
generating a dense point cloud of the mapped environment (in
V. M ETHODOLOGY .LAS format). This can be done by the means of WebODM.
A. Methodology Workflow After that, the dense point cloud is uploaded into GRASS
GIS where the DEM of the mapped region is created and
In this paper, an efficient workflow is presented in order to saved as .ASCII format. Then using Matlab the .ASCII file is
generate the optimal path of the UAV. The workflow is shown converted into .mat file. The aforementioned procedures were
in Figure 1 below.. not performed in this work because of the non-availability of
the UAV system. The .mat file is used in Matlab in order to
B. UAV generate the optimal path for the UAV from a starting location
Any UAV can be used for obtaining high resolution over- to a final location using the forward dynamic programming
lapping images by the means of a camera attached to it and explained in the previous sections of this paper. At the end,
set at the nadir angle. Before doing so, a flight planning has a series of waypoints can be extracted and sent to the UAV
to be set. Flight plan is a waypoint navigation process used to system in order to execute the generated optimal path.
achieve the desire map of the Area of Interest AOI. Generally,
aerial photographs should be overlapped at least 60% forward
lap and 30% side lap. In addition to this, the UAV is used to VI. R ESULTS A ND D ISCUSSIONS
execute the optimal path by flying with constant translational The simulation took place in Windows 8 environment
and rotational speeds. installed in an HP Pavilion branded computer with Intel®
Core™ i3, 4,00 Gb of RAM and 64 bits operating system.
C. Data In our case, the time reduction was moderate; nevertheless
For this work, the imagery data are not available. However, it turns to be a decisive criterion in other applications like
DEMs may be built using Matlab. In our case, the DEM map search and rescue. Since the photogrammetry (time consuming
is created using the Diamond Square Algorithm available in process) was skipped, the simulation lasted for few minutes.
[16]. The Matlab function tends to create raster data of the The first result was creating the DEM using Matlab in a form
map (in .mat format) that are similar to the DEM obtained
from the photogrammetry process. of matrix where each element refers to the height at that point.

Authorized licensed use limited to: London School of Economics & Political Science. Downloaded on May 17,2021 at 14:21:43 UTC from IEEE Xplore. Restrictions apply.
 2020 International Conference on Electrical Engineering (ICEE) September 25-27, 2020, Istanbul, Turkey

The DEM was plotted and then displayed as shown in Figure

2 and Figure 3.

Fig. 4. The optimal path for the first simulation in 2D.

Fig. 2. The generated DEM in 2D.

Fig. 5. The optimal path for the first simulation in 3D.

Fig. 3. The generated DEM in 3D.

After that, we uploaded the DEM generated into the forward

dynamic programming Matlab code and identified both the
starting location and final location. Finally, we run the code.
As a result, the optimal path was generated as shown in
the following figures. In Figure 4 and Figure 5, the starting
location and target location were selected to be (5.0, 3.0) and
(35.0, 61.0), respectively. However for Figure 6 and Figure 7,
they were selected to be (60.0, 10.0) and (20.0, 63.0)

As it can be seen in the above figures, the proposed

approach in this work solves the path planning problem over a
DEM for UAVs. The forward dynamic programming generates
Fig. 6. The optimal path for the second simulation in 2D.
efficiently complete optimal paths from any starting location
to any final location on the DEM. However, the path generated
consists of straight lines and sharp turns. Such path may
contradict with the UAV kinematics since it cannot make To make the generated path desirable, smoothing it represent
suddenly those sharp turns and requires decreasing its speed. a crucial step before sending any commands to the UAV. This

Authorized licensed use limited to: London School of Economics & Political Science. Downloaded on May 17,2021 at 14:21:43 UTC from IEEE Xplore. Restrictions apply.
 2020 International Conference on Electrical Engineering (ICEE) September 25-27, 2020, Istanbul, Turkey

R EFERENCES
[1] L. Yang, J. Qi, J. Xiao, X. Yong. A Literature Review of UAV 3D
Path Planning The 11th World Congress on Intelligent Control and
Automation (WCICA ’14), pp. 2376–2381, 2014.
[2] J. Gruener, G. Siegwart, R. Weingarten. A State-Of-The-Art 3D Sensor
For Robot Navigation. Proceedings of the IEEE/RSJ International
Conference on Intelligent Robots and Systems, pp. 2155–2160, 2004
[3] H. Eisenbeiss. UAV photogrammetry. Dissertation. ETH No. 18515,
Institute of Geodesy and Photogrammetry, ETH Zurich, Switzerland,
Mitteilungen Nr.105, p235, 2009
[4] H. Eisenbeiss. UAV Photogrammetry in Plant Sciences and Geology. In:
6th ARIDA Workshop on ”Innovations in 3D Measurement, Modeling
and Visualization, Povo (Trento), Italy, 2008
[5] S. Shekhar, H. Xiong. Encyclopedia of GIS. Springer, New York. 2008
[6] I. Hasircioglu, H. R. Topcuoglu, M. Ermis. 3-D Path Planning For
The Navigation of Unmanned Aerial Vehicles by Using Evolutionary
Algorithms. In Proceedings of the Conference on Genetic and
Evolutionary Computation, pages 1499-1506, 2008.
[7] R. Bin Omar. Path Planning For Unmanned Aerial Vehicles Using
Fig. 7. The optimal path for the second simulation in 3D. Visibility Line-Based Methods. Ph.D Thesis. University of Leicester,
2011.
[8] Y0 Li, H. Yao, H. Yu, H. Fenghua, Y. Yao An Improved APF-based Path
Planning Algorithm for a Quadrotor Intercepting Autonomous Ground
Robots. Chinese Control Conference (CCC), pp. 4772-4777, 2019.
can be done by using one of the techniques known in literature [9] A. Majeed, S. Lee. A Fast Global Flight Path Planning Algorithm Based
on Space Circumscription and Sparse Visibility Graph for Unmanned
like polynomial interpolation, Bézier curve, cubic splines or by Aerial Vehicle. Electronics, 7(12), 375, 2018.
solving an optimization problem which minimizes parameters [10] M. Kothari, I. Postlethwaite, D. Gu. Multi-UAV Path Planning in
like acceleration, jerk, snap and so on. As a result, the Obstacle Rich Environments Using Rapidly-exploring Random Trees.
Proceedings of the IEEE Conference on Decision and Control, pp. 3069-
path becomes smoother and shorter. This may increases the 3074, 2009.
endurance of the UAV. [11] F. Leonardo, L. Medeiros, J. D. S. Da Silva. A Dijkstra Algorithm for
FixedWing UAV Motion Planning Based on Terrain Elevation. Springer-
Verlag Berlin Heidelberg, 213-222, 2010.
The proposed approach can take into account the static ob- [12] V. Jeauneau, L. Jouanneau, A. Kotenkoff. Path Planner Methods for
stacles present in the environment. Hence, very high resolution UAVs in Real Environment. International Federation of Automatic
photogrammetric DEMs is needed and the UAV is required Control IFAC, pp. 292-297, 2018.
[13] M. Cafaro, I. Epicoco, M. Pulimeno. Techniques for Designing Bioinfor-
to fly with very low altitude, in this case. Nevertheless, the matics Algorithms. Encyclopedia of Bioinformatics and Computational
approach does not take into account the dynamic obstacles Biology, 5-14, 2019.
present in the environment since the photogrammetry process [14] A. Sitinjak , E. Pasaribu , J. E. Simarmata , T. Putra , H. Mawengkang.
The Analysis of Forward and Backward Dynamic Programming for
does not map moving objects. However, there is a possibility Multistage Graph. Conf. Series: Materials Science and Engineering 300,
that such obstacles may be present in the mapped environment. 012010, 2017.
Hence, the path replanning in required. This can be performed [15] A. Mpkrane, A. Choukchou-Braham, B. Cherki. DEM Generation Based
On UAV Photogrammetry. Conference: 2019 International Conference
by a sensor placed on the UAV system. This latter uses the on Advanced Electrical Engineering (ICAEE), 2019.
sensor information to plan a new path locally until the obstacle [16] 3D Fractal Landscapes using the Diamond
is cleared. Square Algorithm. https://cis.temple.edu/ lakam-
per/courses/cis350 2004/assignments/assignment 02.htm. (accessed
10/02/2020)

VII. C ONCLUSION

This work presented a new method for generating an offline

point-to-point optimal path for an UAV using three software:
WebODM, GRASS GIS and MatLAB. The aim of WebODM
was to provide a point cloud of the environment where
the UAV navigated. GRASS GIS software was utilized for
generating a high resolution rasterized DEM from the point
cloud. Afterwards, a forward dynamic programming algorithm
was implemented in Matlab and used divide the optimal
path problem into local subproblems, compute a matrix cost
function over the DEM then generate an optimal path between
a starting and target configurations. The simulation results
showed that this approach is applicable to flight path planning
for any aerial vehicle.

Authorized licensed use limited to: London School of Economics & Political Science. Downloaded on May 17,2021 at 14:21:43 UTC from IEEE Xplore. Restrictions apply.