A Concept for Fast Indoor Mapping and Positioning in Post-Disaster

Scenarios

Eduard Angelats and José A. Navarro

Centre Tecnològic de Telecomunicacions de Catalunya (CTTC/CERCA),

Av. Carl Friedrich Gauss, 7. Building B4, 08860 Castelldefels, Spain

Keywords: RPAS Photogrammetry, Rapid Indoor Mapping, Indoor Positioning, 3d Modelling, Emergency Response,

Orientation, RGB-D Camera.

Abstract: This work presents an early concept for a low-cost, lightweight fast mapping and positioning system suitable

for civil protection and emergency teams working in post-disaster scenarios. Such a concept envisages

continuous, seamless tracking in both indoor and outdoor environments; this would be possible because of

the low geometric requirements set by emergency teams: knowing what is the floor and room where they are

located is enough. The authors believe that current technologies (both hardware and software) are powerful

enough to build such system; such an opinion is backed by an assessment of several currently available sensors

(IMUs, RGB-D cameras, GNSS receivers), embedded processors and mapping and positioning algorithms as

well as a feasibility study taking into account the several factors playing a role in the problem.

1 INTRODUCTION

The ultimate goal of the concept presented in this

paper is to increase the safety of civil protection and

emergency (CPE) teams working in post-disaster

(either natural or man-made) scenarios, as

earthquakes or fires. The members of these teams are

constantly exposed to situations that may put their

lives at (even death) risk. Such risk is significantly

increased due to the lack of knowledge of the

environment they work in - typically, confined spaces

as damaged buildings.

Increasing such knowledge should have a direct

impact on their security. Here, "knowledge" refers to

information about the places (either outdoors or

indoors) where these teams work. There exist, indeed,

solutions to first map an area and then track people in

booth indoor and outdoor enviroments. In the case of

outdoors, there are nowadays many companies

operating Remotely Piloted Aircraft Systems (RPAS)

producing high quality cartography very quickly.

Note, for example, that even the Copernicus

Emergency Management Service (CEMS) offers a

fast mapping service for emergencies using either

Sentinel imagery or RPAS with a response time of

about 48 hours – (CEMS, 2015); locating someone

oudoors is also routinely performed: common

solutions rely on Global Navigation Satellite Systems

(GNSS) receivers or an hybridization of these with

Inertial Measuring Unit (IMU) sensors. There also

exist solutions for positioning in confined places, but

these rely on pre-deployed infrastructure (as WiFi

emitters/beacons or cameras, among others) that will

not be available when working in post-disaster

scenarios since these may be located anywhere.

Solutions for indoor mapping based usually on Light

Detection and Ranging (LiDAR) sensors carried

either by humans or even by terrestrial robots exist

(Kruijff-Korbayová, I. 2016), but these are not always

suitable to operate in post-disaster scenarios; walls

may have collapsed, debris or holes may impede their

normal operation and, in the case of human carriers,

operating in such situations is, doubtlessly,

dangerous.

Thus, it is possible to say that, in practice, a

suitable solution to track seamlessly the position of a

CPE team when moving from outdoors to indoors - or

viceversa - does not exist. As stated above, it is true

that solutions in general do exist, but their

dependance on some infrastructures make these

unsuitable for emergency and disaster management.

On the other side, it is the authors' belief that using

the current technologies and algorithms it is possible

to develop a system able to solve precisely this

problem for the specific case of CPE teams.

274

Angelats, E. and Navarro, J.

A Concept for Fast Indoor Mapping and Positioning in Post-Disaster Scenarios.

DOI: 10.5220/0006780102740281

In Proceedings of the 4th International Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2018), pages 274-281

ISBN: 978-989-758-294-3

To focus the discussion aftewards, it is worth to

note that outdoor mapping and positioning is nothing

new nowadays. High levels of accuracy and precision

far exceeding the requirements of CPE teams are

achieved routinely. Therefore, and even though this

paper presents a seamless solution for mapping and

positioning in both indoor and outdoor environments,

the outdoors case will not be discussed here from the

mapping standpoint. Regarding positioning, the

problem is solved too, but here a concept for a

positioning device able to switch between outdoors

and indoors conditions will be presented.

Such concept targets at a solution to overcome the

difficulties stated above, relying on (1) a low-cost,

lightweight, unobtrusive, portable device (2) to be

carried as a payload by a special RPAS to map indoor

environments and (3) to be carried afterwards by CPE

members when inspecting the buildings. Note the

double role played by the aforementioned device:

mapper and tracker.

The sensors used to build such a device would be Red,

Green, Blue and Depth (RGB-D) cameras, IMU

sensors and GNSS receivers. Data fusion algorithms

would be the heart of the system. These would

combine raw observations coming from the IMU and

the visual odometry solution estimated from RGB-D

cameras’ imagery to provide indoor positioning. The

addition of a GNSS receiver immedately enables such

device for outdoor environments. All these

components (both hardware and software) should be

mounted on a light, battery powered System-On-Chip

(SOC) computer. Depending on the purpose

(mapping or positioning), a different algorithm would

be used.

Concerning the RPAS, not all of them would be

appropriate for the task of mapping the interior of

damaged buildings. Fixed-wing ones should be

immediately discarded for obvious reasons. But even

multi-copters may be severely damaged in case of

crash against a wall. There exist, however, RPAS

designed especifically to resist blows and crashes.

These have been successfuly used for other purposes

where flying in confined spaces was a must. See, for

instance (Flyability, 2017).

This paper will try to show that it is already

possible to build a low-cost fast 3D mapping and

positioning system targeted at the specific needs of

CPE teams using technology and algorithms already

available. One of the factors making this possible is

the low geometric (accuracy and precision) set by the

regular operation of CPE personnel; basically, their

members want to be sure about the floor and room

they are in. This would translate to accuracies around

1-2 metres and precisions between 30-50 cm. To do

it, sensors and algorithms will be presented and

assessed together with the drawbacks derived from

the specific work conditions usually present in the

target scenarios - as suboptimal lightning, or presence

of dust, usually found indoors in many kinds of

natural or man-made disasters. The architecture of the

mapping and positioning system will also be

discussed.

2 THE STATE OF THE ART

2.1 Latest Trends in RPAS Mapping

According to (Gomez and Purdie, 2016; Colomina

and Molina, 2014; Nex and Remondino, 2014), the

number of publications referring to RPAS as a

mapping tool has increased exponentially, reflecting

the rapid spread of this technology. Such tendency

played an important role in the significant reduction

of price undergone by this equipment. Additionally,

(Colomina and Molina, 2014) state that the

improvement of automation software together with

the decrease of price of processors and positioning

and remote sensing sensors are responsible of the

broad use of RPAS in the mapping arena. RPAS have

become a suitable tool for mapping, because of their

ability to carry a variety of mapping and positioning

payloads, as cameras, LiDAR, IMUs or GNSS

receivers (Giordan, 2017), so they may be used to

support systems targeted at emergency management

or hazard assessment. This may be done at local or

even regional scales (Giordan, 2017), which depends

on the kind of RPAS used, that is, fixed-wing or

multi-copter ones (Gomez and Purdie, 2016;

Boccardo, 2015). Last, but not least, one of the

advantages of using and RPAS as a mapping platform

is that these may be deployed very quickly. This

means that RPAS are very suitable for mapping land

features and, especially, their evolution over time or

after sudden, violent transformations as flooding or

volcanic eruptions. A non-exhaustive review of

RPAS mapping for in the context of natural disasters

may be found in (Gomez and Purdie, 2016).

Mapping emergency scenarios with the help of

RPAS is usual when producing outdoors cartography.

This technique is appreciated and valued by CPE and

Search & Rescue (SAR) teams (Boccardo, 2015;

Kruijff-Korbayová, 2016; Dominici, 2017;

Scaramuzza, 2014). On the contrary, RPAS are

almost unknown when working in indoor

environments. Some research projects are the

exception to the previous statement; (Kruijff, 2012;

Kruijff-Korbayová, 2016), in the context of the

A Concept for Fast Indoor Mapping and Positioning in Post-Disaster Scenarios

275

projects NIfTi and TRADR, describe the use of aerial,

RGB imagery to assess the damages suffered by

buildings during the earthquakes that took place in

Emilia-Romagna (2012) and Umbria (2016). 3D

models were built in post-processing mode to perform

the task. Leaving the area of emergency management,

other mapping research projects exist. One of these is

the work of (Mur-Artal and Tardós, 2017) where a

variant of the Simultaneous Localization and

Mapping technique (SLAM) called ORB-SLAM is

presented.

With regard to algorithms and tools, a robust,

widely used technique to produce 3D models using

imagery obtained from Commercial Off-The-Shelf

(COTS) cameras mounted on RPAS is the so-called

Structure-from-Motion (SfM) approach (Nex and

Remondino, 2014). One of the advantage of using the

SfM approach is lower costs, especially when

compared to those incurred when using an RPAS and

a LiDAR sensor. Another advantage is the

availability of a variety of highly automated tools

such as Micmac (Pinte, 2017), Agisoft (Agisoft,

2017) or Pix4D (Pix4D, 2017). (Remondino, 2017)

shows that using GPU-enabled, modern processors

the time to collect and process data is noticeably

reduced, thus leading to processing times in the range

of a few hours, which fully aligns with the goal of

quickly producing quality cartography when

emergency management is the business.

LiDAR measurements may be combined with the

positions or the trajectory to derive, in real-time either

3D models or Digital Surface Models (DSM). The

geometric quality of the overlapping point cloud

strips is directly influenced by the errors that may be

present in the positions in the trajectory. To improve

the quality of both the absolute and relative

orientations and, consequently, that of the point

clouds registration, a post-processing step is proposed

in (Glira, 2016)

2.2 The State of the Art in Positioning

and Mapping

In spite of being a tool widely used lately by most of

the applications needing precise and robust

positioning, the well-known GNSS have some

drawbacks, such as the need of good environment

conditions, as clear lines of sight; confined spaces or

deep canyons (either natural or urban) are the typical

environments where GNSS receivers are not the best

technology for achieving precise positioning results.

As already mentioned in section 1 and in the specific

case of indoor environments, this limitation is usually

overcome (Dardari, 2015) by means of the ad-hoc

deployment of different kinds of emitters (as Wi-Fi,

ultra-wideband or even visual beacons) that are

complemented with the corresponding receiver,

which, aided by the appropriate algorithms is able to

estimate a solution. The emitters play the role of

landmarks - also known as anchor nodes - since these

have been deployed in known positions. This is not,

however, the only approach used indoors.

(Leutenegger, 2013; Veth, 2011) describe the use of

combined IMU data and other sensors (as monocular,

stereo or RGB-D cameras).

When using imaging sensors, it is possible to

compute their orientation parameters by measuring

tie-points in consecutive images that are conveniently

overlapped. The extraction, description and matching

of these tie-points have been the subject of quite a

number of research works proposing several

algorithms performing the aforementioned tasks in a

robust way. Some of these (Lowe, 1999;

Leutenegger, 2011) are, ideally, invariant to changes

in illuminations conditions, orientation and scale.

RANdom SAmple Consensus (RANSAC)

procedures are usually combined with these

algorithms in order to detect and remove outliers; this

is done (Hartley and Schaffalitzky, 2004; Nister,

2003) using only image observations by means of

position and attitude estimation. (Veth, 2011; Taylor,

2011) combine, instead, inertial data and derived

trajectories to predict the appearance of already

detected features in new images.

Removing outliers, it is possible to rebuild the

trajectory by means of the concatenation of the

estimated relative positions and orientations using k

inliers from overlapping images. (Hartley, 2000;

Scaramuzza, 2011; Forster, 2017) review several

methods to estimate the navigation states using

images with or without extra object observations. The

robotics and computer vision communities refer to

these approaches as SfM or visual odometry. These

are called SLAM when drifts are reduced by means

of reference maps or loop closures are detected.

Another line of though (Taylor, 2011), present two

strategies, both using an Unscented Kalman Filter and

an IMU as the main positioning sensor; its drift is

controlled by means of visual information. The first

approach use image coordinates to set geometric

constraints while the second estimates the navigation

states at the same time than object coordinates.

GISTAM 2018 - 4th International Conference on Geographical Information Systems Theory, Applications and Management

276

3 THE CONCEPT

3.1 User Requirement and Scenarios

The concept for the system presented in this paper is

oriented at helping the CPE teams to manage and

assess post-disaster situations, especially in those

cases where the damages inflicted to buildings is

severe enough as to put the lives of the personnel

having to intervene at risk of losing their lives. Not all

the disaster scenarios are targets for this concept, but

at least the following are: volcanic eruptions,

landslides, earthquakes, fires, flooding, severe

storms. In all cases, the system would be used once

the emergency is over.

The proposed system (concept) would not be an

independent one; it would complement the

technologies and procedures that already exist, thus

improving - or trying to improve - the management of

emergencies, including the plans devised to handle

these. From the temporal standpoint, such a system

would be used during the intervention phase, that is,

once the emergency itself (fire, earthquake or any

other of the aforementioned situations) has finished.

The idea, is to increase the safety of the members of

the CPE teams, as well as to improve the resources

available to the coordination personnel thanks to an

assessment of the damages suffered by the buildings

where these people have to work (for instance, being

able to tell where the risk of collapsing walls is

greater). This is an indirect way to say that the goal is

reduce the risk these teams are exposed to, risks that

sometimes take its toll on human lives. An extra

benefit obtained from such a system would be the

ability to help in building rehabilitation (thanks to the

3D models obtained during the intervention phase).

The system should be able to map the disaster area

in only a few hours, both outdoors and indoors.

Outdoors quality RPAS mapping is a service offered

by many companies as a regular product. Among

these, the Copernicus Emergency Management

Service; therefore, it will not be discussed here.

But when talking about indoors and emergency

management specifically, neither the positioning nor

the mapping issues are well solved. In fact, many

problems may affect the ability to create indoor maps

in these environments. The authors, aware of these

difficulties, restrict the scope of applicability of their

work only to those cases in which the lightning and

texture conditions are suitable for the kind of sensors

integrating the system - namely, RGB-D cameras.

Thus, when possible, the system would produce

3D models of the damaged buildings, so the CPE

teams know what they should expect when entering

there. 2D floorplans are another possible output. An

interesting discussion arising here is the quality level

needed to produce these models; unlike other

applications that rely on strict accuracy and precision

requirements, emergency teams typically need just to

be aware of their surroundings - that is, whether a

dangerous wall that could collapse is around, if a

staircase leading up or down is available or if the floor

they should step on still exists, for instance - and what

is the room and floor they are in. Such relaxed

requirements make possible the concept to map and

afterwards track personnel indoors described in this

work. Nonetheless, the authors consider that

minimum accuracy and precision requirements

should be stated in order to assess the suitability of

the system: 3 to 5 decimetres for precision and 1 to 2

meters for accuracy. Such requirements, moreover,

allow for a representation of the buildings close

enough to reality - so it is useful.

The availability of the 3D models and 2D

floorplans opens the way to track the CPE teams and

pinpoint their positions once they enter the buildings.

This information (position) combined with the now

available knowledge about the environment (holes,

collapsed walls, availability of emergency exits, etc.)

produces invaluable information for the member of

both intervening and coordination teams. Again,

some precision requirements should be set for the

indoor positioning: below 10 meters. It is not

possible, however, to assess accuracy since no

reliable reference to compare to exist. Positions

should be updated at least once per second (1 Hz) to

effectively track the teams.

3.2 The Hardware

The system devised by the authors to implement both

the mapping and tracking devices (or mapping

payload and portable positioning device,

respectively) will sport the same hardware

components. A battery powered SOC board will be

used to run the required algorithms (either positioning

or mapping) and to integrate the required sensors.

These will be and RGB-D camera plus an integrated

GNSS / IMU module. This last module could be the

ublox NEO-M8U (ublox, 2017) or a similar one. It

has been chosen because its ability to work in two

different modes; when a GNSS signal is available, it

delivers a position with a frequency of 2 Hz (twice the

required frequency). This would be the typical

outdoor scenario, where GNSS is, normally, not a

problem. But when moving indoors, this module

switches to the untethered dead reckoning mode and

delivers linear accelerations and angular velocities

A Concept for Fast Indoor Mapping and Positioning in Post-Disaster Scenarios

277

instead of position information - that is, the IMU data.

The frequency is 20 Hz. This behaviour matches the

operational mode that should be implemented in the

mapping and tracking devices.

With regard to cameras, it is worth to note that the

evolution of this technology has produced active

RGB-D cameras able to take depth measurements in

extreme lightning conditions - that is, 0 lux or no light

at all. In such cases, the operating distance is

drastically reduced, so, from the operational

standpoint, data should be captured at much closer

ranges. It is true that the accuracy and precision are

also significantly worse; the key point here, however,

is that using one of such active cameras instead of

passive ones makes possible to operate the system

when the illumination and texture requirements set in

section 3.1 are not matched. Obviously, the complete

absence of light (the 0-lux condition) may be too

extreme for a pilot to be able to fly the RPAS; albeit,

the idea is that even when the light is very poor, the

system may be operated and results that may help to

save lives, be obtained. Two cameras already

available in the market would be good candidates for

this system: the Intel© RealsenseTM (Keselman et

al., 2017) and Microsoft Kinect 2 (Lachat et al.,

2015). Unfortunately, the Kinect must be discarded

because of excessive power consumption reasons

(section 4 presents a feasibility analysis, and power

consumption is one of the factors to take into

account).

Finally, an unobtrusive SOC (also with low power

requirements) would be desirable to complete (and

integrate) the set of components making the system.

Its task, to provide with the necessary computing

resources - positioning version - and storage capacity

(mapping version, sensor data must be saved). The

word unobtrusive above means lightweight and small

footprint in this context, since the positioning device

must not be a nuisance to its wearers. Obviously, the

low consumption requirements lead to longer

operational times, thus reducing the need to replace

batteries so often. From the computing power

standpoint, a powerful GPU is a must, due to the

image processing computations that must take place

in real time. A possible candidate for the SOC is the

NVIDIA Jetson TX2 (Franklin, 2017).

3.3 Operating the System

From the operational standpoint, using the system in

indoor environments implies going through three

main steps, namely data collection, map generation

and actual intervention.

First of all, data must be collected using the

mapping payload on board the RPAS. As already

stated in section 3.2, the integrated GNSS + IMU

module will provide GNSS-based data while

operating outdoors, so the RPAS position and attitude

will be computed relying on this information. The

camera plays no role at this stage (since the goal of

this procedure is to produce indoor cartography). It is

important to collect this information since it will be

used as (presumably good) initial approximations for

the attitude and position of the RPAS when it enters

the building and the GNSS signal disappears. As soon

as this happens, the system will activate the RGB-D

sensor and the GNSS + IMU module will start

delivering linear accelerations and angular velocities.

It should not be forgotten that the imagery and IMU

data stored to later produce the 3D models (and 2D

floorplans) must be correctly time-tagged. When

leaving the building, the GNSS-based operations

mode is reactivated, so, again, more precise position

and attitude data will be available, helping to improve

the final quality of the mapping process. The RPAS

itself deserves a few words; usual RPAS are not

suited to operate in confined spaces because of the

high risk of crashing against the walls or any debris

that may be present inside the building. Such crashes

or blows may damage - and therefore, disable - the

RPAS itself. To overcome such difficulties,

multicopters adapted to fly in such circumstances

exist. Flyability's (Flyability, 2017) is one example of

RPAS used successfully in projects requiring flying

indoors.

Once the drone exists the building, it is necessary

to download the data and compute the 3D models and

2D floorplans. The usual software workflow and tools

used for such task are described in section 3.4. The

time needed to obtain these products will directly

depend on the dimensions of the building(s) to

process, although current software is pretty fast and

able to deliver results in short times. However, it must

be noted that RGB-D cameras may deliver a coarse

point cloud immediately. Since it has not been

processed, the accuracy and precision of such data

will never be as good as that of the products obtained

after a post-processing step; nonetheless, and in very

critical situations, the availability of a first point cloud

in so short a time may prove vital for the CPE teams,

being possible for them to enter a building with first-

hand knowledge about it. When there are lives at a

stake, coarse point clouds may make the difference.

Then, the CPE teams, carrying the portable

positioning device, will proceed to enter the already

mapped buildings. Their position will be computed in

real-time by means of visual odometry plus IMU data.

GISTAM 2018 - 4th International Conference on Geographical Information Systems Theory, Applications and Management

278

Ideally, this position should be sent to the control

team outside the building so they can track the

position of the personnel. This, obviously, implies the

use of some kind of communications that will not be

described in this paper. Note that, depending on the

conditions indoors (dust, no lightning, smoke, etc.)

the portable device might be unable to compute

positions. Even in this case, the 3D models and 2D

floor maps in the hands of the control team are

invaluable tools: these may be used to guide the teams

inside the building using classic communication

channels - for example, telling them where to find

holes, collapsed ones, debris, or any other obstacles

that might interfere in their task.

3.4 About the Software

Although the portable positioning device and the

mapping payload share the same SOC and sensors,

the software used to manage it will be different

depending on how it is used.

When used as a portable positioning device, two

situations may be told apart: indoors and outdoors

positioning. The last of these two cases is well known

and usually solved by means of GNSS receivers; in

the concept presented here, IMU data will be used to

enhance position and attitude data. The well-known

extended Kalman filter - or sequential least squares

algorithm - is the engine to use to do so (Parés and

Colomina, 2015). Data will be referred to a global

reference frame. But as soon as the emergency team

enters a building, thus being indoors, the device will

change its mode of operation, thus using a different

algorithm. Since the GNSS signal will not be

available, the RGB-D cameras plus the IMU raw data

will be used instead. Again, an extended Kalman

filter will be used. The IMUs will play the main role

in the prediction step, producing orientation and

position data in a global reference frame. On the

contrary, and during the filter step, the orientation and

position data derived from the processing of the

RGB-D images (visual odometry) will be responsible

for updating the predicted ones, in order to control the

drift introduced by the inertial sensors. In this step,

data is referred to a local reference frame, but it is

possible to compute their equivalent in the global one

using the level arm and boresight matrices. Finally, a

common temporal reference frame is necessary to

correctly deal with data coming from these two

sources - the internal clock of the SOC will be enough

for the purposes of this concept.

When the device is used as a mapping payload,

it is targeted at the production of the 3D models and

2D floorplans. Once that the data (GNSS -outdoors-,

IMU - indoors + outdoors - and RGB-D images -

indoors) have been downloaded, three steps are

necessary to do it.

In first place, it is necessary to estimate the

(coarse) initial approximations of the positions and

attitude values related to the imagery. The algorithm

is the same used by the device when working as a

portable positioning device (see above). Secondly, a

block adjustment will take care of refining these

initial approximations so much better values are

obtained. Depending on the community, this step is

known as SfM or "Integrated Sensor Orientation".

This task may be done using any of the software

packages available nowadays, as Pix4D (Pix4D,

2017), AgiSoft (AgiSoft, 2017) or Micmac (Pinte,

2017). Finally, the depth data obtained from the

RGB-D camera together with the position and

orientation data just refined will be used to produce a

dense point cloud. A popular software library able to

do this is PCL (PCL, 2017).

4 FEASIBILITY

Going from a concept to an actual, working

implementation safely, implies a feasibility analysis.

Some factors that must be taking into account in such

analysis have a direct impact on the hardware use to

implement the positioning / mapping device.

Power consumption. Some RGB-cameras are

pretty greedy regarding power. Obviously, higher

power consumption implies less operating time and

the unsuitability of the camera for the purposes of this

work. This is the case of the Microsoft Kinect v2.

Extra hardware or software requirements. Some

cameras, including again the Kinect v2, require extra

hardware to work properly and to deliver the

performance needed to fulfil the pursued goals

(Chesa, 2017). Examples of these extra requirements

are GPUs, OpenGL, or USB 3.0 ports.

SOC performance. Several authors have tested the

suggested SOC, the NVIDIA Jetson TX2, showing

that it is capable enough to produce the necessary

output rate in terms of positions per second when

using the algorithms and techniques described in this

paper (Mur-Artal and Tardós, 2017; Forster, 2017).

Common temporal frame. The positioning /

mapping device collects data originating in different

sensors (the RGB-D camera and the GNSS + IMU

module). The algorithms used to produce positions or

maps out of these observations need to know the

precise time when these observations were generated,

and such time must be referred to a common temporal

frame for all the sensors involved in the process.

A Concept for Fast Indoor Mapping and Positioning in Post-Disaster Scenarios

279

Systems having to match high-quality geometric

constraints normally use devoted hardware to provide

the aforementioned temporal reference frame. In the

case discussed here, the SOC's internal clock is

enough, specially (although not necessarily) if the

drifts and latencies existing in the system are properly

characterized.

Pre-heating time, calibration and RGB-D data

quality. An incorrect (or inexistent) geometric

calibration or too short pre-heating times may

produce systematic errors in the quality of the depth

measurements that have direct implications in that of

the derived point cloud (Chesa, 2017; Keselman,

2017; Lachat 2015). It is worth to note that the

geometric requirements in the context of this

application are so relaxed that the aforementioned

problems are negligible in this case.

Environmental operating conditions. Visual

odometry needs some minimum operating conditions

to work, especially those regarding illumination and

texture. The operating range as well as the quality of

the depth measurements directly depend on them. The

immediate consequence is that the quality of the

positioning / mapping solution, relying on the

extraction of features of the RGB-D imagery, is also

affected. (Mur-Artal and Tardós, 2017; Forster, 2017)

discuss how feasible is to operate in moderate

illumination conditions. Note, also, that the when

using active sensors (infrared emission) it is possible

to work in near 0-lux lightning conditions, although

the operating range is still shorter.

Mapping post-processing times. According to

(Remondino, 2017), the photogrammetric software

packages currently available are powerful enough as

to deliver results in times short enough as to meet the

requirements set by post-disaster, emergency

scenarios. Furthermore, the limited autonomy of

RPAS still reinforces the previous statement, since

the amount of data to process will be relatively small,

thanks to the capabilities of the aforementioned

software packages and current computers.

RGB-D cameras' extra cost. RGB-D cameras are

slightly more expensive than RGB ones but,

according to (Mur-Artal and Tardós, 2017; Fang,

2015) the extra features these incorporate are worth

of it. RGB-D cameras are still able to produce robust

solutions in poor or changing illumination conditions.

Finally, the point cloud these cameras provide may be

improved (see previous point above) by means of

post-processing techniques, delivering useable 3D

models in very short times.

5 CONCLUSIONS

This work was motivated by the risk that CPE teams

take whenever they enter damaged buildings in post-

disaster scenarios. The main goal was to check

whether a fast indoors mapping / positioning system,

relying on no pre-deployed infrastructures, and

delivering 3D models with a quality enough to fulfil

the needs of that teams, could be feasible using

currently available technologies and recent advances

in algorithms.

A thorough state of the art has been presented,

showing the most recent technology and algorithms;

a concept, relying heavily on that information, has

been detailed, explaining how such a system could be

built and exploited. A feasibility analysis,

highlighting the most relevant issues has also been

included.

This paper presents just a concept and not an

actual implementation of the system. Nonetheless, the

performances, features, algorithms and procedures

already explored and documented by many other

researchers have convinced the authors that such a

system is feasible nowadays. And that it is possible

using low-cost equipment, which would facilitate its

adoption by many civil protection agencies.

REFERENCES

Agisoft, 2017. Agisoft Photoscan. http://www.agisoft.com.

Accessed: 30 November 2017.

Boccardo, P., F. Chiabrando, F. Dutto, F. Giulio Tonolo,

and A. Lingua. 2015. UAV Deployment Exercise for

Mapping Purposes: Evaluation of Emergency Response

Applications. Sensors 15: 15717–15737.

CEMS, 2015. Copernicus Emergency Management Service

Mapping. Manual of Operational Procedures

Guidelines for EC Services, Service Providers and

Authorized Users. European Commission DG GROW,

DG ECHO, DG JRC. Version 1.1 February 2015.

Chesa, M., 2017. Obstacle avoidance for an autonomous

Rover. Bachelor degree thesis Technical University of

Catalonia, 2017.

Colomina, I., and P. Molina. 2014. Unmanned aerial

systems for photogrammetry and remote sensing: A

review. ISPRS Journal of Photogrammetry and Remote

Sensing 92: 79–97.

Dardari, D., Closas, P., Djuric, P, 2015. Indoor Tracking:

Theory, Methods, and Technologies. IEEE

Transactions on Vehicular Technology, 64(4), 2015,

1263-1278.

Dominici, D., Alicandro, M., Massimi, V., 2017. UAV

photogrammetry in the post-earthquake scenario: case

studies in L'Aquila. Geomatics, Natural Hazards and

Risk Vol. 8, Iss. 1, 2017.

GISTAM 2018 - 4th International Conference on Geographical Information Systems Theory, Applications and Management

280

Fang, Z.; Zhang, Y., 2015. Experimental Evaluation of

RGB-D Visual Odometry Methods. International

Journal of Advanced Robotics Systems 2015, 12, 1–16.

Flyability, 2017. ELIOS - Inspect & explore indoor and

confined spaces. http://www.flyability.com/elios.

Accessed: 30 November 2017.

Franklin, D., 2017. NVIDIA Jetson TX2 delivers twice the

intelligence to the edge. https://devblogs.nvidia.com/

parallelforall/jetson-tx2-delivers-twice-intelligence-

edge. Accessed: 30 November 2017.

Forster, C., Zhang, Z., Gassner, M., Werlberger, M.,

Scaramuzza, D., 2017. SVO: Semi-Direct Visual

Odometry for Monocular and Multi-Camera Systems.

IEEE Transactions on Robotics, Vol. 33, Issue 2, pages

249-265, Apr. 2017.

Giordan, D., Manconi, A., Remondino, F. and Nex, F.C.,

2017. Use of unmanned aerial vehicles in monitoring

application and management of natural hazards: open

access. Geomatics, Natural Hazards and Risk, 8

(2017)1 pp. 1-4.

Glira, P., Pfeifer, N., Mandlburger, G., 2016. Rigorous Strip

Adjustment of UAV-based Laserscanning Data

Including Time-Dependent Correction of Trajectory

Errors. Photogrammetric Engineering and Remote

Sensing, 82 (2016), 12; 945 – 954.

Gomez, C., Purdie, H., 2016. UAV- based Photogrammetry

and Geocomputing for Hazards and Disaster Risk

Monitoring – A Review. Geoenvironmental Disasters,

2016, Volume 3, Number 1, Page 1.

Hartley, R. and Zisserman, A., 2000. Multiple view

geometry in computer vision. Cambridge University

Press, 2nd edition.

Hartley, R. and Schaffalitzky, F., 2004. Minimization in

geometric reconstruction problems. In: Proceedings of

Computer Vision and Pattern Recognition conference,

2004.

Keselman, L., Iselin Woodfill, J., Grunnet-Jepsen, A.,

depth cameras. In: Proceedings of IEEE Conference on

computer vision and pattern recognition workshops

(CVPRW), 21-26 July 2017, Honolulu, HI (USA).

Kruijff, G.-J., Tretyakov, V., Linder, T., Pirri, F, Gianni,

M., Papadakis, P., Pizzoli, M., Sinha, A., Pianese, E.,

Corrao, S., Priori, F., Febrini, S. and Angeletti, S., 2012.

Rescue robots at earthquake-hit Mirandola, Italy: A

field report. In IEEE Intl. Symp. on Safety, Security, and

Rescue Robotics (SSRR), 2012.

Kruijff-Korbayová, I. 2016. The Use of Robots for Disaster

Response”. Presentation at CTIF Delegates Assembly

& Symposium, Sep 8-9 2016 in Helsinki.

Lachat, E., Macher, H., Landes, T., Grussenmeyer, P.,

2015. Assessment and calibration of a RGB-D Camera

(Kinect v2 sensor) towards a potential use for close-

range 3D modeling. Remote Sensing 7(10):13070-

13097.

Leutenegger, S., Chli, M., Siegwart, R., 2011. BRISK:

Binary Robust Invariant Scalable Keypoints. In:

Proceedings of the IEEE International Conference on

Computer Vision (ICCV), 2011.

Leutenegger, S., Furgale, P.T., Rabaud, V., Chli, M.,

Konolige, K. and Siegwart, R., 2013. Keyframe-Based

Visual-Inertial SLAM using Nonlinear Optimization,

In: Proceedings of Robotics: Science and Systems,

2013.

Lowe, D., 1999. Object recognition from local scale-

invariant features. In: Proceedings of the International

Conference on Computer Vision (ICCV), 2011.

Mur-Artal R., and Tardós, J.D, 2017. ORB-SLAM2: an

Open-Source SLAM System for Monocular, Stereo and

RGB-D Cameras. IEEE Transactions on Robotics, vol.

33, no. 5, pp. 1255-1262, June 2017.

Nex, F.C., and Remondino, F., 2014. UAV for 3D mapping

applications: a review. Applied Geomatics, 6 (2014)1

pp. 1-15.

Nister, D., 2003. An efficient solution to the five-point

relative pose problem. In CVPR03.

Parés, M.E., Colomina, I., 2015. On software Architecture

Concepts for a Unified, Generic and Extensible

Trajectory Determination System. In: Proceedings of

the ION GNSS+, 08-12 September 2015, Tampa,

Florida (USA).

PCL., 2017. Point Cloud Library (PCL).

http://pointclouds.org/. Accessed: 7 February 2018.

Pinte, A., 2017. Micmac, un logiciel pour la mise en

correspondance automatique dans le contexte

géographique. http://logiciels.ign.fr/?Micmac.

Accessed: 30 November 2017.

Pix4D, 2017. Pix4D - Drone photogrammetric software for

desktop + cloud + mobile. https://pix4d.com.

Accessed: 30 November 2017.

Remondino, F., Nocerino, E., Toschi, I., and Menna, F,

2017. A Critical Review of Automated

Photogrammetric processing of large datasets. In: Int.

Arch. Photogramm. Remote Sens. Spatial Inf. Sci.,

XLII-2/W5, 591-599, 2017.

Scaramuzza, D. and Fraundorfer, F., 2011. Visual

Odometry [Tutorial]. IEEE Robot. Automat. Mag.

18(4): 80‐ 92.

Scaramuzza, D. et al., 2014. Vision-Controlled Micro

Flying Robots: From System Design to Autonomous

Navigation and Mapping in GPS-Denied

Environments. IEEE Robotics & Automation

Magazine, vol. 21, no. 3, pp. 26-40, Sept. 2014.

Taylor C.N, Veth M.J., Raquet J.F., Miller M.M, 2011.

Comparison of Two Image and Inertial Sensor Fusion

Techniques for Navigation in Unmapped

Environments. IEEE Transactions on Aerospace and

Electronic Systems, vol. 47, no. 2, pp. 946–958.

ublox, 2017. NEO-M8U module. https://www.u-blox.com/

en/product/neo-m8u-module. Accessed: 30 November

2017.

Veth, M.J., 2011. Navigation Using Images, A survey of

Techniques. Journal of Navigation 58(2), pp. 127–140.

A Concept for Fast Indoor Mapping and Positioning in Post-Disaster Scenarios

281