Designing Transparent and Autonomous

Intelligent Vision Systems

Joanna Isabelle Olszewska

School of Engineering and Computing, University of West Scotland, U.K.

Keywords:

Intelligent Vision Systems, Autonomous Systems, Vision Agents, Agent-oriented Software Engineering,

Explainable Artiﬁcial Intelligence (XAI), Transparency and Ethical Issues.

Abstract:

To process vast amounts of visual data such as images, videos, etc. in an automatic and computationally

efﬁcient way, intelligent vision systems have been developed over the last three decades. However, with

the increasing development of complex technologies like companion robots which require advanced machine

vision capabilities and, on the other hand, the growing attention to data security and privacy, the design of

intelligent vision systems faces new challenges such as autonomy and transparency. Hence, in this paper,

we propose to deﬁne the main requirements for the new generation of intelligent vision systems (IVS) we

demonstrated in a prototype.

1 INTRODUCTION

With the omnipresence of digital data in our Society,

and in particular visual data (Olszewska, 2018) from

smartphone pictures to television video streams, from

m-health services to social media apps, from street

surveillance cameras to airport e-gates, from drones

to autonomous underwater vehicles (AUVs), intelli-

gent vision systems (IVS) are needed to automatize

the processing of these visual data.

Indeed, intelligent vision systems are a set of in-

terconnected hardware and/or software components

which take digital image(s) as input data and process

them by means of methods ranging from low- to high-

level techniques/algorithms in order to extract mean-

ingful information, which could be structured and or-

ganized into knowledge, and aid to the automatic un-

derstanding of the gathered visual data (Fig.1).

The input image to be processed by an IVS could

be a single, still picture, a set of pictures, or a dy-

namic stream of frames (Sabour et al., 2008). The

image(s) could be gathered from an online/ofﬂine

database (Rahbi et al., 2016) or acquired live by a

single vision sensor or multiple ones (Bianchi and

Rillo, 1996), each sensor being either static or dy-

namic (Ishiguro et al., 1993).

The IVS output is the information obtained after

processing the input visual data. The resulting in-

formation could present any degree of modality, i.e.

could be of a semantic type (degree 0) such as a

tag/label or a text ﬁle, of a visual type itself, e.g. a

picture or a region of a picture (degree 1), or of a

video type (degree 2). This information could also

consist in further processed results such as computed

trajectories (Ukita and Matsuyama, 2002), workﬂows

(Sardis et al., 2010), etc. It could produce knowledge

(Reichard, 2004) and/or provide further understand-

ing of the visual data (Li et al., 2009).

In order to process the input visual data, a (series

of) method(s) is implemented within the intelligent

vision system. It could consist of low-level techniques

of image processing such as thresholding, morpho-

logical operations, edge detection, texture region de-

tection (Arbuckle and Beetz, 1999), etc. Mid-level

techniques include computer vision methods such as

interest point descriptors, active contours (Olszewska,

2015), etc. High-level techniques involve artiﬁcial in-

telligence (AI), and in particular, machine learning

approaches which could be based on symbolism (e.g.

logic rules) (Olszewska, 2017), analogism (e.g. Sup-

port Vector Machine (SVM)) (Prakash et al., 2012),

probability (e.g. Bayesian rule) (Hou et al., 2014),

evolutionarism (e.g. genetic algorithms) (Nestinger

and Cheng, 2010), or connectivism (e.g. neural net-

works) (Jeon et al., 2018).

The design of such complex systems necessitates

a careful requirement analysis (Rash et al., 2006), not

only in terms of performance targets of the vision al-

gorithm(s), but also in terms of software and system

requirements.

850

Olszewska, J.

Designing Transparent and Autonomous Intelligent Vision Systems.

DOI: 10.5220/0007585208500856

In Proceedings of the 11th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2019), pages 850-856

ISBN: 978-989-758-350-6

Figure 1: Intelligent vision system (IVS) overview.

Indeed, in this paper, we present the requirements

for intelligent vision systems.

IVS can be considered either as an intelligent vi-

sion software or an embodiment of a vision process.

Therefore, the proposed set of requirements pro-

vides the backbone for the ethical and dependable de-

sign of such complex vision systems.

The main contributions of this work are, on one

hand, the identiﬁcation of the requirements for the

new generation of intelligent vision systems and, on

the other hand, the deﬁnition of the main concepts

such as autonomy and transparency in context of in-

telligent vision systems as well as the elucidation of

the intelligent vision system notion itself and its re-

lated design pattern.

The paper is structured as follows. In Section 2,

we present the intelligent vision system requirements

and their related deﬁnitions. The proposed design

method has been successfully prototyped as reported

and discussed in Section 3. Conclusions are drawn up

in Section 4.

2 REQUIREMENTS

The proposed IVS requirements are intelligent vision

system’s reliability, security, autonomy, and trans-

parency, as described in Sections 2.1-2.4, respec-

tively.

2.1 Reliable

The prime focus of the IVS design has been the ef-

ﬁciency of such systems in order to develop reliable

solutions.

Low- and mid-level IVS performance are assessed

using one or more metrics quantifying shape ﬁdelity

(Correia and Pereira, 2003), shape accuracy (Gelasca

and Ebrahimi, 2009), shape temporal coherence (Er-

dem and Sankur, 2000), connectivity, and compact-

ness (Goumeidane and Khamadja, 2010).

However, there is no single deﬁnition of these

measures. In particular, the segmentation error could

be deﬁned in several ways, e.g. by calculating:

• the Pratt’s Figure of Merit (FOM) (Pratt et al.,

1978) which evaluates the edge location accuracy

via the displacement of detected edge points from

an ideal edge:

FOM =

max(I

)

∑

i=1

1 + δ d

(i)

, (1)

with I

, the number of ideal edge points, I

, the

number of actual edge points, d, the Euclidean

distance between the actual points and the ideal

edge points, and δ, a scaling constant (of usual

value δ = 1/9). The value of FOM falls between

0 and 1, and the larger, the better.

• the overlap (OL) between the segmented region A

and the ground truth region A

(Saeed and Duge-

lay, 2010):

OL =

2(A∩)A

A + A

∗ 100, (2)

where OL ∈ [0,100], and the larger, the better.

• the segmentation error (SE) for edge and region-

based methods (Saeed and Dugelay, 2010):

SE =

2 ∗ A

∗ 100, (3)

where SE ∈ [0,100], and the smaller, the better.

• the spatial accuracy or MPEG-4 Segmentation er-

ror Se (Muller-Schneiders et al., 2005):

Se =

∑

k=1

∑

l=1

card(M

)

, (4)

where card(M

) is the number of pixels of the ref-

erence mask; fn, the number of false negative pix-

els; fp, the number of false positive pixels; d

, the

distance of the k

false negative pixel to the ref-

erence pixel; and d

, the distance of the l

false

Designing Transparent and Autonomous Intelligent Vision Systems

851

Figure 2: Representation of the true positive rate (TP), the

false positive rate (FP), the false negative rate (FN), and the

true negative rate (TN), respectively, in a confusion matrix.

positive pixel to the reference pixel. The value of

Se is in the range of [0,∞[, and the smaller, the

better.

The above-mentioned measures (Eqs. 1-4) could

lead to further assessment of IVS, e.g. by plotting

detected/tracked object’s actual trajectory vs ground

truth one (Leibe et al., 2008).

The effectiveness of mid- and high-level IVS is

mainly measured using the following set of metrics

(Estrada and Jepson, 2005):

precision (P) =

T P

T P + FP

, (5)

detection rate (DR) or (R) =

T P

T P + FN

, (6)

f alse detection rate (FAR) =

FP + T P

, (7)

accuracy (ACC) =

T P + T N

T P + T N + FP + FN

, (8)

where TP is the True positive rate, FP is the False

Positive rate, FN is the False Negative rate, and T N

is the True Negative rate (see Fig. 2).

It is worth noting the detection rate (DR) is also

sometimes called recall (R), sensitivity, or hit rate.

Another common metric is the F1 score which is

the harmonic mean of the precision and recall and

which could be used when a balance between preci-

sion and recall is needed and when the class distri-

bution is uneven (i.e. high T N + FP). F1 score is

deﬁned as follows:

F1 score (F1) = 2

P ∗ R

P + R

. (9)

To represent and evaluate IVS efﬁciency, the stan-

dard measures (Eqs. 5-8) could be used to compose a

confusion matrix such as in Fig. 2 and/or a precision-

recall curve. The later one is useful in the case the

classes are very imbalanced and shows the trade-off

between precision and recall for different thresholds.

It is worth noting a high area under the precision-

recall curve is aimed, since it represents both high

recall (i.e. low FN) and high precision (i.e. low FP).

Indeed, high scores for both precision and recall show

that the system is returning accurate results (high pre-

cision) as well as returning a majority of all positive

results (high recall), i.e. the system returns many re-

sults, with most of the results labeled correctly. Inci-

dentally, a system with high recall but low precision

returns many results, but most of its predicted labels

are incorrect when compared to the training labels. A

system with high precision but low recall returns very

few results, but most of its predicted labels are correct

when compared to the training labels.

On the other hand, programming paradigms and

implementation language choices have an impact on

the IVS design, with C++/OpenCV and MatLab the

most common languages adopted for IVS (Nestinger

and Cheng, 2010). Moreover, the development of re-

liable IVS is a Test-Driven Development (TDD) that

can be assessed using software engineering indica-

tors (Pressman, 2010) such as function-based met-

rics, speciﬁcation quality metrics, architectural design

metrics, class-oriented metrics, component-level de-

sign metrics (including cohesion and coupling met-

rics), operation-oriented metrics, user-interface de-

sign metrics, etc. Hence, IVS code complexity, main-

tainability, and quality (Reichardt et al., 2018) as well

as IVS computational speed (e.g. computational time

vs image resolution/size) are crucial to be analysed in

the IVS design phase. .

Furthermore, IVS robustness and fault tolerance

(Asmare et al., 2012) along with its portability and

interoperability (Bayat et al., 2016) should also be

considered in the IVS design phase, as IVS tend to

be deployed not only on desktops/laptops, but also on

smartphones, robots, industrial equipment, etc.

2.2 Secure

In today’s society, ensuring the security of cyber-

physical systems is a main challenge (Escudero et al.,

2018), (Burzio et al., 2018). Therefore, in addition of

correctness and robustness (Reichard, 2004), IVS de-

sign should integrate the cybersecurity element (Rus-

sell et al., 2015). For example, the communication

between the visual data acquisition set-up and the ma-

chine processing them should be secured (Leonard

et al., 2017).

Furthermore, the collected visual data which are

ICAART 2019 - 11th International Conference on Agents and Artiﬁcial Intelligence

852

Figure 3: UML Class Diagram of the Intelligent Vision System.

processed by IVS as well as the produced informa-

tion and knowledge should respect data privacy and

in particular comply with GDPR legislation (General

Data Protection Regulation (GDPR), 2018).

Hence, the design of such correct, robust, and se-

cure IVS ensures its dependability (Meyer, 2006).

2.3 Autonomous

Autonomous systems (AS) are systems that decide

themselves what to do and when to do it (Fisher et al.,

2013). Their features could include failure diagnosis,

self-awareness, biomimetic, automated reasoning or

knowledge-inspired transactions (Olszewska and Al-

lison, 2018).

IVS autonomy is the system’s capacity of manag-

ing itself using artiﬁcial intelligence (AI) method(s)

to produce the intended goal(s), i.e. to process

and provide the expected information, to build the

relevant knowledge, and/or to automatically anal-

yse/understand its environment. Depending of its

level of autonomy (i.e. not autonomous, semi-

autonomous, autonomous), the system will have a

various degree of interactions with other intelligent

agents (Calzado et al., 2018) through manned or re-

mote control operations (Zhang et al., 2017).

Figure 3 presents a possible pattern for such

systems, whatever their classiﬁcation (Franklin and

Graesser, 1996), architecture (Fiorini et al., 2017), or

environment (Osorio et al., 2010).

2.4 Transparent

Transparency in AS is the property which makes pos-

sible to discover how and why the system made a par-

ticular decision or acted the way it did (Chatila et al.,

2017), taking into account its environment (Lakhmani

et al., 2016).

IVS transparency is the system’s capacity of its

goals, its situational constraints, its input/output data,

its decision criteria, its internal structure, its assump-

tions about the external reality, its actions, and its in-

teractions being understood by the relevant stakehold-

ers. Hence, depending of the level of transparency,

the system could be transparent to the system’s users,

commanders, regulators, and/or investigators.

Therefore, an IVS system should adopt the de-

sign pattern proposed in Fig. 3. Moreover, designing

transparent IVS implies the choice of intelligent tech-

niques, like the machine learning (ML) methods, not

only in terms of efﬁciency but also in terms of trans-

parency. As per (Bostrom and Yudkowsky, 2014), the

Designing Transparent and Autonomous Intelligent Vision Systems

853

most transparent ML techniques are logic based and

the less transparent ones involve neural networks. In-

deed, with explainable AI (XAI) (Ha et al., 2018) be-

ing in its infancy, logic-based approaches (Olszewska

and McCluskey, 2011) have the most well-established

procedures for system veriﬁcation (Brutzman et al.,

2012). In addition, ML training databases should

avoid biases (Skirpan and Yeh, 2017).

3 PROTOTYPE

The proposed requirements have been demonstrated

on a IVS prototype aiming to detect and count peo-

ple in indoor environments (Fig. 4). The system can

be used by a university to monitor the use of com-

puter labs during assignment deadline times. This can

help a university to determine whether the computer

labs should be open for longer during these periods.

Another beneﬁt of the developed application is that it

can be used for safety purposes to provide an accurate

number of people inside a building, aiding the evacu-

ation process e.g. in the event of a ﬁre.

The IVS prototype has been run on a computer

with Intel Core i5-2400 CPU, 3.10 GHz, 12Gb RAM,

64-bit Windows 7 Enterprise OS, using MatLab lan-

guage.

As per Section 2, the IVS reliability has been as-

sessed in terms of recall (R = 80%) and precision

(P = 91%). The computational speed rate is 14 fps.

The IVS security has been analysed and the data

privacy has been ensured in order people appearing

within the video will remain anonymous.

The IVS system has been designed in order to

work in an autonomous way, performing the tasks as

follows:

• Detecting people - the software is able to detect

the people in the video automatically.

• Tracking people - the software automatically

tracks people in the video, once they are detected.

• Counting people - the software is able to automat-

ically count the people detected in the video.

The IVS prototype has been designed in order

to be transparent to both users and experts. Indeed,

from the user’s point of view, the input for this soft-

ware is the video-recorded stream that the software

processes, and the software output is in the form of

two video players that show simultaneously the an-

notated input and the extracted information all along

the software execution (Fig. 4). From the expert’s

point of view, the software process does not involve

deep learning and is rather based on the main intel-

ligent vision techniques such as background subtrac-

Figure 4: Example of the intelligent vision system proto-

type in action.

tion and foreground detection by thresholding, blob

analysis by applying morphological mathematic op-

erations, tracking using a Kalman ﬁlter (KF), and a

counter algorithm to count the people detected; all

these techniques being formally veriﬁable.

4 CONCLUSIONS

In this work, we proposed a set of requirements and

deﬁnitions to aid the design of (near-)future intelli-

gent vision systems (IVS). Indeed, intelligent vision

systems have been developed for 30 years, with a

main focus on their efﬁciency. Nowadays, IVS pres-

ence is ever growing in our daily life. Hence, their

design should not only take into account their relia-

bility in terms of accuracy and robustness, but also

encompass concepts such as transparency, autonomy,

security, and privacy. Experiments in real-world con-

text have displayed the effectiveness and usefulness of

our approach well-suited for applications within im-

age/video processing, pattern recognition, computer

vision, and machine vision domains.

ACKNOWLEDGEMENTS

The author would like to thank her students, R. Dobie

and J. Webb for their involvement in the prototype.

REFERENCES

Arbuckle, T. and Beetz, M. (1999). Controlling image pro-

cessing: Providing extensible, run-time conﬁgurable

functionality on autonomous robots. In Proceedings

of the IEEE/RSJ International Conference on Intelli-

gent Robots and Systems (IROS), pages 787–792.

Asmare, E., Gopalan, A., Sloman, M., Dulay, N., and Lupu,

E. (2012). Self-management framework for mobile

autonomous systems. Journal of Network and Systems

Management, 20(2):244–275.

ICAART 2019 - 11th International Conference on Agents and Artiﬁcial Intelligence

854

Bayat, B., Bermejo-Alonso, J., Carbonera, J. L.,

Facchinetti, T., Fiorini, S., Goncalves, P., Jorge, V.

A. M., Habib, M., Khamis, A., Melo, K., Nguyen, B.,

Olszewska, J. I., Paull, L., Prestes, E., Ragavan, V.,

Saeedi, S., Sanz, R., Seto, M., Spencer, B., Vosughi,

A., and Li, H. (2016). Requirements for building an

ontology for autonomous robots. Industrial Robot: An

International Journal, 43(5):469–480.

Bianchi, R. A. C. and Rillo, A. H. R. C. (1996). A purpo-

sive computer vision system: A multi-agent approach.

In Proceedings of the IEEE Workshop on Cybernetic

Vision, pages 225–230.

Bostrom, N. and Yudkowsky, E. (2014). The ethics of arti-

ﬁcial intelligence. The Cambridge Handbook of Arti-

ﬁcial Intelligence.

Brutzman, D., McGhee, R., and Davis, D. (2012). An im-

plemented universal mission controller with run time

ethics checking for autonomous unmanned vehicles

— a uuv example. In Proceedings of the IEEE Inter-

national Conference on IEEE/OES Autonomous Un-

derwater Vehicles, pages 1–8.

Burzio, G., Cordella, G. F., Colajanni, M., Marchetti, M.,

and Stabili, D. (2018). Cybersecurity of connected

autonomous vehicles : A ranking based approach. In

Proceedings of the IEEE International Conference of

Electrical and Electronic Technologies for Automo-

tive, pages 1–6.

Calzado, J., Lindsay, A., Chen, C., Samuels, G., and Ol-

szewska, J. I. (2018). SAMI: Interactive, Multi-Sense

Robot Architecture. In Proceedings of the IEEE Inter-

national Conference on Intelligent Engineering Sys-

tems, pages 317–322.

Chatila, R., Firth-Butterﬂied, K., Havens, J. C., and

Karachalios, K. (2017). The IEEE global initiative

for ethical considerations in artiﬁcial intelligence and

autonomous systems. IEEE Robotics and Automation

Magazine, 24(1):110.

Correia, P. L. and Pereira, F. (2003). Objective evaluation

of video segmentation quality. IEEE Transactions on

Image Processing, 12(2):186–200.

Erdem, C. E. and Sankur, B. (2000). Performance evalua-

tion metrics for object-based video segmentation. In

Proceedings of European Signal Processing Confer-

ence (EUSPICO), pages 1–4.

Escudero, C., Sicard, F., and Zamai, E. (2018). Process-

aware model based idss for industrial control sys-

tems cybersecurity: Approaches, limits and further re-

search. In Proceedings of the IEEE International Con-

ference on Emerging Technologies and Factory Au-

tomation (ETFA), pages 605–612.

Estrada, F. J. and Jepson, A. D. (2005). Quantitative eval-

uation of a novel image segmentation algorithm. In

Proceedings of the Joint IEEE Computer Society Con-

ference on Computer Vision and Pattern Recognition

(CVPR), pages 1132–1139.

Fiorini, S. R., Bermejo-Alonso, J., Goncalves, P., de Fre-

itas, E. P., Alarcos, A. O., Olszewska, J. I., Prestes, E.,

Schlenoff, C., Ragavan, S. V., Redﬁeld, S., Spencer,

B., and Li, H. (2017). A suite of ontologies for

robotics and automation. IEEE Robotics and Automa-

tion Magazine, 24(1):8–11.

Fisher, M., Dennis, L., and Webster, M. (2013). Verifying

autonomous systems. Communications of the ACM,

5-6(9):84–93.

Franklin, S. and Graesser, A. (1996). Is it an agent, or just

a program?: A taxonomy for autonomous agents. In

Proceedings of the Intelligent Agents III Agent The-

ories, Architectures, and Language, LNCS, Springer,

pages 21–35.

Gelasca, E. D. and Ebrahimi, T. (2009). On evaluating video

object segmentation quality: A perceptually driven

objective metric. IEEE Journal of Selected Topics in

Signal Processing, 3(2):319–335.

General Data Protection Regulation (GDPR) (2018). Regu-

lation (EU) 2016/679 of the European Parliament and

of the Council of 27 April 2016 on the protection of

natural persons with regard to the processing of per-

sonal data and on the free movement of such data.

Goumeidane, A. B. and Khamadja, M. (2010). Error mea-

sures for segmentation results: Evaluation on syn-

thetic images. In Proceedings of IEEE Interna-

tional Conference on Electronics, Circuits and Sys-

tems, pages 158–161.

Ha, T., Lee, S., and Kim, S. (2018). Designing explainabil-

ity of an artiﬁcial intelligence system. In Proceedings

of the ACM Conference on Technology, Mind, and So-

ciety.

Hou, Y., Zhang, Y., Xue, F., Zheng, M., and Fan, R. (2014).

The fusion method of improvement Bayes for soccer

robot vision system. In Proceedings of IEEE Interna-

tional Symposium on Computer, Consumer and Con-

trol, pages 776–780.

Ishiguro, H., Kato, K., and Tsuji, S. (1993). Multiple vision

agents navigating a mobile robot in a real world. In

Proceedings of the IEEE International Conference on

Robotics and Automation (ICRA), pages 772–777.

Jeon, H.-S., Kum, D.-S., and Jeong, W.-Y. (2018). Traf-

ﬁc scene prediction via deep learning: Introduction of

multi-channel occupancy grid map as a scene repre-

sentation. In Proceedings of IEEE Intelligent Vehicles

Symposium (IV), pages 1496–1501.

Lakhmani, S., Abich, J., Barber, D., and Chen, J. (2016).

A proposed approach for determining the inﬂuence of

multimodal robot-of-human transparency information

on human-agent teams. In Proceedings of the Inter-

national Conference on Augmented Cognition, pages

296–307.

Leibe, B., Schindler, K., Cornelis, N., and Gool, L. V.

(2008). Coupled object detection and tracking from

static cameras and moving vehicles. IEEE Transac-

tions on Pattern Analysis and Machine Intelligence,

30(10):1683–1698.

Leonard, S. R., Allison, I. K., and Olszewska, J. I. (2017).

Design and Test (D & T) of an in-ﬂight entertainment

system with camera modiﬁcation. In Proceedings of

the IEEE International Conference on Intelligent En-

gineering Systems, pages 151–156.

Li, L.-J., Socher, R., and Fei-Fei, L. (2009). Towards total

scene understanding: Classiﬁcation, annotation and

segmentation in an automatic framework. In Proceed-

ings of the IEEE International Conference on Com-

Designing Transparent and Autonomous Intelligent Vision Systems

855

puter Vision and Pattern Recognition (CVPR), pages

2036–2043.

Meyer, B. (2006). Dependable software. In Kohlas, J.,

Meyer, B., and Schiper, A., editors, Dependable Sys-

tems, pages 1–33. Springer.

Muller-Schneiders, S., Jager, T., Loos, H. S., and Niem, W.

(2005). Performance evaluation of a real time video

surveillance system. In Proceedings of the Joint IEEE

International Workshop on Visual Surveillance and

Performance Evaluation of Tracking and Surveillance

(PETS), pages 137–143.

Nestinger, S. S. and Cheng, H. H. (2010). Flexible vision.

IEEE Robotics and Automation Magazine, 17(3):66–

77.

Olszewska, J. (2017). Detecting hidden objects using ef-

ﬁcient spatio-temporal knowledge representation. In

Revised Selected Papers of the International Confer-

ence on Agents and Artiﬁcial Intelligence. Lecture

Notes in Computer Science, volume 10162, pages

302–313.

Olszewska, J. I. (2015). Active contour based optical char-

acter recognition for automated scene understanding.

Neurocomputing, 161:65–71.

Olszewska, J. I. (2018). Discover intelligent vision systems.

In DDD Scotland 2018.

Olszewska, J. I. and Allison, I. K. (2018). ODYSSEY: Soft-

ware Development Life Cycle Ontology. In Proceed-

ings of the International Conference on Knowledge

Engineering and Ontology Development, pages 301–

309.

Olszewska, J. I. and McCluskey, T. L. (2011). Ontology-

coupled active contours for dynamic video scene un-

derstanding. In Proceedings of the IEEE International

Conference on Intelligent Engineering Systems, pages

369–374.

Osorio, F., Wolf, D., Branco, K. C., and Pessin, G. (2010).

Mobile robots design and implementation: From vir-

tual simulation to real robots. In Proceedings of ID-

MME 2010, pages 1–6.

Prakash, J. S., Vignesh, K. A., Ashok, C., and Adithyan,

R. (2012). Multi class support vector machines clas-

siﬁer for machine vision application. In Proceedings

of IEEE International Conference on Machine Vision

and Image Processing, pages 197–199.

Pratt, W., Faugeras, O. D., and Gagalowicz, A. (1978). Vi-

sual discrimination of stochastic texture ﬁelds. IEEE

Transactions on Systems, Man, and Cybernetics,

8(11):796–804.

Pressman, R. S. (2010). Product metrics. In Software en-

gineering: A practitioner’s approach, 7th Ed., pages

613–643. McGraw-Hill.

Rahbi, M. S. A., Edirisinghe, E., and Fatima, S.

(2016). Multi-agent based framework for person re-

identiﬁcation in video surveillance. In Proceedings

of the IEEE Future Technologies Conference, pages

1349–1352.

Rash, J. L., Hinchey, M. G., Rouff, C. A., Gracanin, D., and

Erickson, J. (2006). A requirements-based program-

ming approach to developing a NASA autonomous

ground control system. Artiﬁcial Intelligence Review,

25(4):285–297.

Reichard, K. M. (2004). Integrating self-health awareness

in autonomous systems. Robotics and Autonomous

Systems, 49(1-2):105–112.

Reichardt, M., Foehst, T., and Berns, K. (2018). Introducing

ﬁnroc: A convenient real-time framework for robotics

based on a systematic design approach.

Russell, S., Dewey, D., and Tegmark, M. (2015). Re-

search priorities for robust and beneﬁcial artiﬁcial in-

telligence. AI Magazine, 36(4):105–114.

Sabour, N. A., Faheem, H. M., and Khalifa, M. E. (2008).

Multi-agent based framework for target tracking using

a real time vision system. In Proceedings of the IEEE

International Conference on Computer Engineering

and Systems, pages 355–363.

Saeed, U. and Dugelay, J.-L. (2010). Combining edge de-

tection and region segmentation for lip contour ex-

traction. In Proceedings of International Conference

on Articulated Motion and Deformable Objects, pages

11–20.

Sardis, E., Anagnostopoulos, V., and Varvarigou, T. (2010).

Multi-agent based surveillance of workﬂows. In Pro-

ceedings of the IEEE/WIC/ACM International Confer-

ence on Web Intelligence and Intelligent Agent Tech-

nology, pages 419–422.

Skirpan, M. and Yeh, T. (2017). Designing a moral com-

pass for the future of computer vision using specula-

tive analysis. In Proceedings of the IEEE Conference

on Computer Vision and Pattern Recognition Work-

shops (CVPRW), pages 1368–1377.

Ukita, N. and Matsuyama, T. (2002). Real-time multi-

target tracking by cooperative distributed active vi-

sion agents. In Proceedings of the ACM International

Joint Conference on Autonomous Agents and Multia-

gent Systems (AAMAS), pages 829–838.

Zhang, T., Li, Q., Zhang, C.-S., Liang, H.-W., Li, P., Wang,

T.-M., Li, S., Zhu, Y.-L., and Wu, C. (2017). Current

trends in the development of intelligent unmanned au-

tonomous systems. Frontiers of Information Technol-

ogy and Electronic Engineering, 18(1):68–85.

ICAART 2019 - 11th International Conference on Agents and Artiﬁcial Intelligence

856