Increasing Business Opportunities for Drone Services

S. Brezani

, R. Hrasko

, D. Vanco

and P. Vojtas

1,2 a

Globesy ltd., Framborska 58, Zilina, Slovak Republic

Dpt. Software Engineering, Charles University, Malostranke nam. 25, Prague, Czech Republic

Keywords: UAV Imaging in Business Processes, B2C/B2B Considerations, Drone Services, Machine Learning.

Abstract: We describe our UAV imaging research and development from a business perspective. Our main goal is to

increase business opportunities for our drone services. We build on experiences from an already deployed

drone service provided by our IT company. The application is using a commercial tool for processing

photogrammetry. Its advantage is accuracy, but the main disadvantage is the time needed for annotation by a

trained human operator. Our methodology is based on user studies and knowledge gained in communicating

with potential customers at IT trade fairs and exhibitions. We analyse the duration and automation of the

service as key factors. We consider two types of higher automation of the solution. First is the automation of

annotation - less accurate, without human intervention. The second is automation in flight planning and

implementation. The use of other drone peripherals or hybrid drones can also create new types of services. In

particular, there is a demand for immediate execution of on-site flight imaging without any pre-calibration.

Our considerations, expanding our services, also include various inspections or direct involvement in

industrial processes. Some improvements were tested on an experimental prototype. The results indicate

improvement making services cheaper and faster.

1 INTRODUCTION

We describe our UAV imaging research and

development from a business perspective. Our main

goal and objectives are related to increasing business

opportunities for our drone services.

The author's team consists of the academia and

R&D department of an IT company.

In the beginning, the company has been trying to

gain a new perspective of using drones in commercial

tasks and founded a flight service department.

We quickly realised that we could use drones in

various cases where humans have limited access or

the standard approach takes very long. Following this,

we acquired the conviction that by using the

equipment drones can carry, we can collect a wide

variety of data in an immense amount.

After deciding to follow these ideas, we were able

to finish and market our first service deployment. To

address the further expansion on the market, we are

looking for new opportunities for our services. Our

ambition is to make them cheaper and faster for

greater competitiveness. The use of other drone

https://orcid.org/0000-0002-3526-8475

peripherals can create new types of services as well.

Our considerations, expanding our services, also

include various inspection types or direct

involvement in industrial processes.

The main contributions of our paper are

following:

• experiences learned from our first deployed

service point to critical factor of service duration

• experiments with our experimental tool from the

point of view of time complexity

• further directions of development with different

level of construction and evaluation, namely

− proof of concept of flight planning

automation

− further considerations of extensions based

on experiences

o LIDAR in unknown uncalibrated areas

o drones extended by peripheral devices

o image recognition inspection

The paper is organised as follows: Chapter 2

describes our starting point with a deployed tool,

lessons learned from customers, and the first

technical attempt to improve it. Chapter 3 describes

Brezani, S., Hrasko, R., Vanco, D. and Vojtas, P.

Increasing Business Opportunities for Drone Services.

DOI: 10.5220/0010566900930100

In Proceedings of the 18th International Conference on e-Business (ICE-B 2021), pages 93-100

ISBN: 978-989-758-527-2

ongoing work, partly, not yet fully validated in

different directions regarding further development

and some speculative ideas. Chapter 4 is related

work. We conclude with achieved results so far, the

future work is devoted to general B2B considerations

about our products' future, especially in the

automotive industry.

2 DEPLOYED UAV SERVICE

We build on experiences our already deployed drone

service developed in our IT company. The customer

requires computing volumes of wood in stock. The

application is using a commercial tool for processing

photogrammetry. The advantage is accuracy, but the

main disadvantage is the need for a trained human

intervention.

Technical description of this solution was

described in (Brezani, 2021). In the following, we

explain how we proceeded, especially in business

considerations.

The product itself is already on the market. It is a

service using a commercial photogrammetry

processing tool for precise measurement and

evaluation. The whole process begins by setting

calibration points, placing them manually, and

marking a known length on-site. Our staff workers

personally create calibration points on the ground

close to the measured area. It makes a base for precise

calibration in the orthophoto map processing method,

which comes later in the procedure.

The next step is to prepare the drone flight data

manually to capture the entire area, focusing on the

material supplies' location. Subsequently, the drone is

ready to fly to take appropriate area pictures. The

flying route is designed depending on the shape of the

object being scanned itself.

Afterwards, photos are taken and collected to fit

automatic processing by a commercial tool, which

creates an orthophoto map. A trained user manually

annotates the areas of interest (wood stock, the

volume should be calculated). It takes a trained user

at least half an hour, depending on the ground area

and object's shape. The procedure ends by

calculating the required material stockpile volume.

The first mission was to estimate the volume of

current material stocks and their changes over time in

the customer warehouse. It took several months for

enough measurements to be made to evaluate the

movement of material in the warehouse.

We focused on several factors that must be

considered separately for specific flight conditions in

different weather conditions when performing

measurements. Each scan and measurement brought

various issues. Meteorological conditions influence

flights to a large extent. It is necessary to modify

predefined plans for flight safety assurance and to

create an optimal setting of the measurement mission

in various conditions.

The main challenge was to reduce the time needed

for the whole procedure and eliminate the manual

interventions and expert labour needed in processing.

The higher requirements on precision require more

processing time in the range of hours to days.

Likewise, manual interventions and professional

work were an obstacle to a significant reduction in

time and full automation of the process. Intelligent

automation was necessary to address processing time.

3 EXPERIMENTAL TOOL

Technological description of our experimental tool is

reported in (Brezani, 2021). The main idea was the

ability to extract information for the customer's

information system. Here we only briefly summarise

the part replacing manual annotation, which is

important for our business considerations. We

decided to use the neural network method of

automated annotation of areas of interest to address

the problem. We used open-source software to create

an orthophoto area map. The orthophoto area map is

then automatically annotated by a neural network

semantically segmenting each pixel of the input

image to see if it belongs to the monitored object. The

before-mentioned network is trained on manually

annotating the initial machine learning examples. It is

important to notice that this annotation is done only

once for the domain). The type of monitoring object

is decided in advance in order to prepare and train a

machine learning model for this task concerning

customer needs. It is time-consuming, but it is a one-

time job, the result of which can be used repeatedly

in services of the same type.

We are using an open-source data labelling,

annotation, and exploration tool to annotate the

images. The goal is to mark the woodpiles (as a pilot

task) on a sample of pictures. Subsequently, we split

the data images into a training dataset, a validation

dataset, and a test dataset. To enlarge the size and

diversity of annotated data, we pick augmentation

within the training cycle to prevent overfitting in

neural networks (Ronneberger, 2015). We were

trying a wide range of transformations to convert both

image and segmentation data. In our case, we are

using affine transformations (rotation, shift, zoom),

contrast adjustment, noise generation, etc.

ICE-B 2021 - 18th International Conference on e-Business

To replace human annotation of the area of

interest, we apply the segmentation models library as

a ground for our work. The library implements four

model architectures for binary and multi-class

classification. It uses the transfer learning method and

allows using one of the pre-trained networks as a

backbone for the semantic segmentation architecture.

The technique makes it possible to use a trained

neural network, or part of it, for related categories of

tasks. For more details, see (Brezani, 2021).

3.1 B2B Considerations

In this chapter, we show the results of our

methodology. Our methodology is based on user

studies and knowledge gained in communicating with

potential customers at IT trade fairs and exhibitions.

Based on this, we analyse the duration and automation

of the service as key factors. We consider two types of

higher automation of the solution. First is the

automation of annotation - less accurate, without

human intervention, and hence faster. The second type

of automation will be treated in the next chapter.

In the previous chapter, we have described the

construction of our experimental tool based on deep

neural network annotation. Here we report on

measuring improvements that were tested on an

experimental prototype. The results indicate

improvement making services cheaper and faster.

Hence this is an indication that we can expect

improvements in goals and objectives.

Nowadays, our ambitions aim to automatically

collect outdoor visual data using pre-programmed

UAVs and automatically process and transform them

into knowledge using advanced computational tools

such as machine learning based on deep neural

networks. Deploying this solution to a real production

facility will bring the capability of automatic data

collection from different devices and their processing

regularly, with possible direct integration to core

information systems in the form of direct data

transfers or, for example, gained knowledge in the

form of alerts. This way, the outdoor reality could be

manageable almost in real-time.

Let's imagine a system that can proceed

automatically without intervention. In that case, we

could streamline the entire process of regular daily

inventory measurements and at the same time

effectively evaluate and monitor the movements and

volume changes of material in the warehouse. We

could potentially get an overview of materials'

movement over large areas of one or more

warehouses of different customers. It would find

justification in many industries by extending today's

limited capabilities to almost unlimited use with

automated drones for regular inventories. Of course,

this is the hope our experimental tool is bearing. It

needs much more work to ascertain and evaluate this

in practical deployment.

Many machine learning tools are distributed as

open-source and are suitable for commercial use.

However, the use of these tools is not trivial and

requires considerable know-how and data. Therefore,

we must include this in the market price of the

instrument. Using open software for machine learning

and processing UAV data, we save on a professional

license for photogrammetry and at the expense of a

trained user. The time saved in terms of speeding up

the procedure will improve our position in the

competition.

In the following section, we describe the effect of

the measured area and the time required to perform

the whole procedure. We also provide further details

here regarding the individual terminations that take

place in the service. These data come from

measurements already performed.

The measurement parameter ranged between 18-

30ha(hectare) for double grid mapping range and

resolution level up to 2 cm/pix. Increasing the

measurement area is directly proportional to the

extension of data processing time.

The diagram in Figure 1 analyses the difference in

the service duration in areas of 100 ha and 30 ha. It

also shows the expected time savings by introducing

automatic and autonomous parts of processes in

several phases. The procedure can be accelerated by

automating manual processes, to which we would be

able to provide results periodically, e.g., every hour.

Thus, it would be possible to continuously control the

increase or decrease of the amount of material.

Customers thus receive a continuous overview of

stockpiles (in the domain of our first deployment)

inventory of materials every hour, every day, month,

and year.

The preparation phase usually lasts 2-4 hours,

depending on the measurement's conditions and nature.

The flight usually takes 40-70 minutes.

The computation phase usually takes 3-10 hours

for maximum quality. The fastest result is within 60

minutes, with an accuracy of around 70%. More

accuracy needs more processing time. The resulting

accuracy can increase to 91-97% over time, but it is

time-consuming. The standard accuracy is around

80%. The required time for the entire procedure with

80% accuracy is, on average, from 5 to 22 hours

together. The following diagram compares each

phase's average duration.

Increasing Business Opportunities for Drone Services

Figure 1: Dependency of processing time, the area, and auto or manual processing shows some huge discrepancies in G. Data

Processing activity which depends on server power. But our focus is mainly on processing steps, which shows the best

improvement by introducing smart technologies, which is apparent in activities B. Data collection planning and H. Objects

of interest annotation. The diagram shows that automatic annotation is one of the key factors to save time and resources.

3.2 Market Situation

Here we summarise the outcomes of our informal

market study. We do not discuss the formal metric

evaluation. On the other hand, this is a continuous

process and depends on the increase of our

experiences.

Several aviation companies are operating in the

market for air inspection services, collecting and

processing aviation data. In our survey, we focused

on providing the maximum added value. Our output

against the competition differs in terms of data

collection to processing and clear presentation of

results. The data collection method's contribution is

more precise thanks to the methodological procedures

and tools used in the solution. Data collection results

are processed by sorting and simple editing of photo

data and then imported into a smart user-friendly

application. The application is available on smart

devices such as mobile phones or tablets. The outputs

are more exact and linked to the location, so the

customer gets accurate data based on which he can

evaluate the current state from a distance. Regular

data import from individual measurements gives

another added value, namely the development of

parameters in time.

The application also contains several tools, thanks

to which it is possible to store status assessment notes

in the photo data. In this way, it is possible to mark

abnormal sites suitable for preventive monitoring.

Regular monitoring helps eliminate power outages by

early detection of the problems in the initial stages.

Finally, our product's descriptive advantages are

access to the databases of individual objects for

assessment. The application is available for Android

and iOS devices, as well as for Windows. In

conclusion of our internal study, the competition does

not provide such benefits in such a form.

4 FURTHER DIRECTIONS

In this chapter, we describe several further directions

of development with different levels of construction.

Some are in the form of proof of concept, and some

ICE-B 2021 - 18th International Conference on e-Business

are speculations (as encouraged by organisers of call

for position papers).

As mentioned in the previous chapter, there is a

second type of increased automation in flight

planning and implementation. Proof of concept gives

us hope that this can contribute to goals and

objectives.

Further, we mention that using other drone

peripherals or hybrid drones can also create new types

of services. In particular, there is a demand for

immediate execution of on-site flight imaging

without any pre-calibration.

Our considerations, expanding our services, also

include various inspections or direct involvement in

industrial processes.

4.1 Flight Planning Automation

In order to shorten the overall duration of the services

and obtain regular and up-to-date data, we proposed

simplifying respective processes so that, as a result,

we can provide repeating services for the recurring

daily needs of the customer. It was necessary to

increase the automation of processes that today take

several to tens of hours and shorten them to minutes.

The process of realising the request for specific data

from the field begins with defining the exact

requirements of the customer. The following is

validated as proof of concept. For this purpose, we

have created a smart application on the web interface

into which the customer directly enters his request,

defines, e.g., the type of data he is interested in, e.g.,

measurement of quantities, map creation,

measurement of warehouses, etc., then enters the

request for the location and the time of execution. It

saves a lot of time in the consultation as the client's

requirements are clearly defined. Of course, it

remains to test the intuitiveness and self-

explainability of our interface.

Based on the request, we are ready to schedule the

autonomous flight so that during it, the drone

performs the necessary data collection. We plan the

flight mission so that the data collection is carried out

safely and so that the necessary data is collected in the

highest possible quality during the mission.

According to several criteria, dependent on the shape

or type of data needed, we set the plan in the flight

planner and adjust the operation of the individual

peripherals. So far, when the flight plan is ready, the

pilot/copilot flight unit drives to the place of activity,

sometimes 100s of km away from the place of the lab.

For this reason, we have designed an automatic

drone box to be placed in an industrial area, from

which the drone is ready to take off and execute the

planned flight within a few minutes. With this

process, we can save a lot of time which can be used

for the implementation of the customer's requirement.

The implementation and data collection will take

place autonomously and under the same conditions

repeatedly. There is no need to plan the flight for each

request. Thus, after the data collection process, we

can import the raw data directly to our remote

processing PC after successful landing by the ground

box. After receiving the data, the metadata check and

verification of correct execution follow. If everything

went well, then the raw data processing into usable

data for the customer as defined in our application

interface follows.

Data processing is a process that takes several

hours in common practice today, depending on the

application. Still, our contribution in the field of

automation and machine learning has pushed the

boundaries and thus shortened this process to the

level of minutes. This is due to the fact that there is

not so much manual work of a trained worker in the

field of post-processing. The software can

automatically recognise and automatically annotate

the object in the scope of interest (as described in the

experimental tool). This way, we can streamline the

processing time by only working with the relevant

data and not processing the rest.

Automating data processing is a key element for

reducing process time and making services cheaper

because it requires minimum manual work.

4.2 Further Extensions

Previous research was evaluated by deployment,

experiments, and proof of concept. In the following

chapter, we describe our further considerations.

4.2.1 LIDAR in Unvisited Uncalibrated

Areas

In the meantime, we have improved our UAVs by

extending our equipment. Our main idea when using

LIDAR is that it will be possible to calibrate the map

without the need for expert intervention.

LIDAR can be an exciting source of data for

automated processing with machine learning. Such as

in self-driving cars - (Sometimes, artificial

intelligence can make better predictions than a human

could because it can access different data, such as

feeds from cameras, RADAR, and LIDAR around a

car (Agrawal, 2019)

LIDAR is a 40 times heavier device than a regular

camera. It is used only in the case of measurements at

the level of cm, where it is necessary to shift

Increasing Business Opportunities for Drone Services

photogrammetry accuracy to a limit of less than 5%.

The LIDAR deviation level is 1-2%, and mapping a

large area is several times faster. Data processing

requirements increase by 20-30 times. The use of

LIDAR is intricate and reduces the drone's flight time,

which increases the cost. It is always necessary to

consider its benefits from shooting with a full-frame

camera.

Besides, this could open up the possibility of

using UAVs in new, unknown areas. We hope this

can attract some customers to use our on-call services,

so we increase our operability. We keep in mind

mountain rescue operations, which cause many

fatalities in our country each year.

To increase the overall efficiency of air services,

we have proposed a procedure for accelerating data

collection through flight automation. In the next step,

it is necessary to adapt the sensors. We know from

practice that it is necessary to obtain a lot of accurate

data for customers. The solution is to use a LIDAR

sensor for mapping large areas over 100ha. Mapping

large areas have great potential for customers, e.g.,

from forestry, agriculture, and automotive.

4.2.2 Inspections by Drones

In other areas of our research and development, we

are working on image recognition. Using this

knowledge can help us use drones in inspection tasks.

So we think of classic inspection tasks, such as power

lines, bridges, etc., see, e.g. (Fowler, K. R. and Dyer,

S. A., 2020). Surprisingly, we do not have much

competition in this area in our country.

Inspections using UAVs are based on detailed

optical data collected from a safe distance using a

20 to 30 fold approximation to specifically defined

locations on the object. During the mast's inspections,

we focus on the insulators' damage, the temperature

of the source adapters and antennas, and the condition

of the construction of the handles with the search for

corrosive joints or screws. Camera correction is

necessary, and the manual flight method is time-

consuming.

The inspection procedure depends on the type of

object. The whole inspection must be pre-planned and

programmed using flight simulation and simulation

of the object's camera views. This way, the entire

infrastructure inspections can be planed. Recognising

and distinguishing objects in the current camera view

is a necessary part of the data collection process.

https://www.techbriefs.com/component/content/article/tb/

pub/features/articles/23938?start=2

4.2.3 Drone Peripheral Extensions

Peripherals are an integral part of drone equipment.

They are selected based on dimensional limits and

parameters that lead to the efficient management of

drone energy. Preferred tools are elementary sensors

and multiple camera types that can capture a wide

range of map layers from a given area, for example,

IR, NIR, RGB, or others.

By combining several data and linking the

measurements of several types of data to a specific

location, we can provide the basis for a detailed, in-

depth analysis of the object, which is fundamental to

the customer's analytical decision-making and

predictive action.

5 RELATED WORK

Market analysts agree that commercial drones are

revolutionising business operations. For example, our

first deployment changed the inventory processes in

the case of our customer. Further summarisation

reads as follows: Drone industry revenue in the

commercial sector is forecast to grow worldwide by a

compound annual growth rate of 13.8 % from 2020 to

2025, reaching a value of 43 billion U.S. dollars,

according to DRONEII. The biggest drone markets

today are in China and Japan. These are not a result

of research papers. These are global market

observations. In this related work, we sometimes give

voice also to subjective opinions of important market

players.

By reading Tech Briefs magazine, it became

evident that UAVs have become an integral part of

the research that uncovers and makes available their

use in various fields. Detection of dangerous

circumstances, such as gas leakage, mapping of

changes in the form and size of glaciers, monitoring

forest stands with early fire detection, rescue

missions, and industrial processes optimisation are

just some of the magazine's reports. An interesting

fact is that UAVs' use is clearly shifting from

manually operated aircraft to autonomous systems

controlled by computer systems involving artificial

intelligence. E.g. (TechBriefs, 2016)

recognised the

need to safely manage UAVs flying at low altitudes

in airspace not currently addressed by authorities.

The American Meteorological Society states in its

research that UAV technology and systems can be

considered a missing piece of the puzzle between

ICE-B 2021 - 18th International Conference on e-Business

satellite observations and observations from the Earth's

surface. AMS states that UAV systems' deployment

will increase the accuracy and timeliness of

meteorological parameters measurements, which will

result in better prediction of weather developments and

abnormalities (see, e.g. (AMS, 2013)

P. Murphy, in his book (Murphy, 2017), ranked

drones and the UAV system among the leading

technologies to move the company to the stage of an

automated company. Interestingly, he identified

transport and means of transport as one of the main

areas for drones (see our future work).

A similar conclusion reached Automotive

Logistics magazine's research, which identified

considerable scope for UAV systems' involvement in

B2B, B2C logistics, and logistics within production

plants and integrated supply chains (see, e.g.

(Williams, 2017)

Nowadays, global trends concern UAV services.

The next part of related work considers cloud and

SaaS, which is not part of our research but the next

step for our customers. So far, the deployed solution

is running on our servers. Nevertheless, in the future,

we have to consider also this option when the

customer decides.

Cloud-based solutions have significantly

increased the availability of sophisticated and

powerful software solutions for research and

economic entities of all sizes. Chue Hong et al., in

their work (Chue Hong, 2018)

, offer a guide for

decision using cloud computing in research. They

warn before too great optimism. One has to check

several questions and dangerous scenarios before

such a decision. Of course, there are also some

benefits possible. (Lakshmi Devasena, 2014) is an

empirical impact study that emphasises the

consequences of adopting Cloud Technology in

business organisations (micro, Small Medium

Businesses (SMBs), and Small Medium Enterprises

(SMEs)) and how it affects business development.

Finally, (Konersmann, 2020)

recognise immense

possibilities cloud computing can offer R&D in Life

sciences and health care organisations in the global

pandemic crisis.

We are at a stage where industrial production is

beginning to open up to the use of SaaS-based

software solutions. After the SaaS model's initial

https://www.ametsoc.org/index.cfm/cwwce/boards/board

-on-enterprise-strategic-topics/offshore-wind-energy-

annual-partnership-topic-committee/apt-final-report/

https://www.automotivelogistics.media/ups-tests-residen

tial-drone-delivery/17665.article

https://www.software.ac.uk/best-practice-using-cloud-

research

change, in which industrial institutions moved

administrative and support information systems to the

cloud, the phase of transition to the SaaS model of

critical production systems begins. Based on research

by Statista, the use of SaaS software in production is

expected to increase by almost 100% by 2020 see

(Statista, 2020)

for 2008 to 2020 data.

There is a more similar material, but we do not

consider it to be mentioned here given the scope. Now

we mention two research papers relevant to our doing.

In the paper (Fotouhi, 2019), the authors study the

rapid growth of consumer unmanned aerial vehicles

(UAVs), creating promising new business

opportunities for cellular operators. UAVs can be

connected to cellular networks as new types of user

equipment, therefore generating significant revenues

for the operators that can guarantee their stringent

service requirements. We are also motivated by this,

as 5G gives enough throughput and makes AI

computations possible on ground computers.

A substantial part of our development is to create

autonomous flying services. In the paper (Jahan,

2019), they consider autonomous systems integrated

into our lives as home assistants, delivery drones, and

driverless cars. The implementation of the level of

automation in these systems from being manually

controlled to fully autonomous would depend upon

the autonomy approach chosen to design these

systems. This is exactly our position. Motivated by

the author's review of the historical evolution of

autonomy, its approaches, and the current trends in

related fields, we incorporate these ideas in our work.

Another option we have to consider for our goals

and objectives is the decision between build and buy.

(Fowler and Dyer, 2020) propose a model for

recommending build-versus-buy decisions when

developing embedded systems. They compare

designing a custom unit with integrating a

commercial unit into the final product (exactly as we

did on our first deployment with commercial

photogrammetry). It accounts for the expertise of the

development team, tool resources available to the

team, partitioning of the tasks, and quality of

commercial units, vendor support, premiums, and

product life cycles. This is now a challenge for our

R&D department. Especially interesting for our flight

department is the paper (Martin, 2018)

mentioning a

https://www2.deloitte.com/us/en/insights/topics/digital-

transformation/cloud-enabled-research-and-

development-innovation.html

https://www.statista.com/statistics/510333/worldwide-

public-cloud-software-as-a-service

https://search.informit.org/doi/10.3316/informit.5911237

71201857

Increasing Business Opportunities for Drone Services

global trend of Asia-Pacific nations making strides

towards supplementing foreign helicopter

acquisitions with domestically built projects. Nations

like South Korea, China, Japan, and India are leading

the way with a number of increasingly sophisticated

transport and utility helicopter designs.

6 CONCLUSIONS

This position paper's main topic is studying the

impact of drones in industrial business and B2B

considerations of its comprehensive application. We

describe ongoing work on further development and

ideas from a business perspective. Our foundation is

in already developed and deployed service using

UAV imaging. We dealt with a further extension of

our service's competitiveness by making it cheaper

and faster and extending services to other domains

like inspections and direct industrial processes

involvement. Some of them are in the phase of

evaluated experiments, some are ascertained as proof

of concept, and some are further considerations based

on our previous experiences.

B2B considerations on the impact of the deployed

tool on the market are well advanced. Drones

extended with LIDAR equipment for unknown

uncalibrated areas offer new market opportunities.

Other ideas discussed and reworked relate to drones

with peripheral enhancements and their use for image

recognition-based control. We evaluate our market

positions and competition. Future ideas include

testing applications to support industrial processes in

polytechnic vocational schools to gain experience in

an environment as close as possible to the real

industrial one. We believe that this will bring us

closer to the significant extensions for drone-based

services. We hope this will help us with our future

research topic for drone-based services in the

automotive industry.

We are trying to create synergy between UAV use

to collect and extract knowledge with immediate

feedback to control and manufacturing processes.

ACKNOWLEDGEMENTS

This publication was realised with the support of the

Slovak Operational Programme Integrated

Infrastructure in the frame of the project: Intelligent

systems for UAV real-time operation and data

processing, code ITMS2014+: 313011V422 and

co-financed by the European Regional Development

Fund.

REFERENCES

Agrawal, A. et al. (2019). Artificial Intelligence: The

Ambiguous Labor Market Impact of Automating

Prediction, J. Economic Perspectives 33,2, pp. 31–50

AMS (2013). American Meteorological Society Offshore

Wind Energy Committee Report

Brezani, S. et al. (2021). Deep Learning for Knowledge

Extraction from UAV Images. Submitted to EJC'21

Chue Hong, N., etal. (2018). Best practice for using cloud

in research, Software Sustainability Institute

Fotouhi, A. et al. (2019). Survey on UAV Cellular

Communications: Practical Aspects, Standardisation

Advancements, Regulation, and Security Challenges.

IEEE Comm. Surveys & Tutorials 21,4, pp. 3417-3442

Fowler, K. R. and Dyer, S. A. (2020). A Model for the

Build-Versus-Buy Decision in Developing Embedded

Systems. IEEE Systems Journal, 14,1, pp. 1317-1324

Jahan, F. et al. (2019). Security Modeling of Autonomous

Systems: A Survey. ACM Comp. Surv. 52,5, Article 91

Konersmann, T., etal. (2020). Innovating R&D with the

cloud Business transformation could require cloud-

enabled data, ecosystems, and services, Deloite insights

Lakshmi Devasena, C. (2014). Impact study of cloud

computing on business development, J. Operations

Research and Applications (ORAJ), 1, 01-07

Martin, G. (2018, March). Build or buy: Imported vs

indigenous transport and utility helicopters. Defence

Review Asia. Asian Press Group.

Murphy, P. (2017). Auto-Industrialism: DIY Capitalism

and the Rise of the Auto-Industrial Society, SAGE

Publications Ltd; 1

edition

Ronneberger, O., et al. (2015). U-Net: Convolutional

Networks for Biomedical Image Segmentation,

https://arxiv.org/abs/1505.04597

Statista, (2020). Total size of the public cloud software as a

service (SaaS) market from 2008 to 2020

TechBriefs, (February 1, 2016). Aerospace Drone Control:

Flying the Crowded Skies,

Williams, M. (2017). UPS tests residential drone delivery,

Automotive Logistics, digital edition

ICE-B 2021 - 18th International Conference on e-Business

100