Virtual Event

Event program

Thursday, 9/30/2021 10:00 AM - 3:00 PM, Online 10:00 AM - 3:00 PM Papers  1. L. Premk, Ž. Emeršič, T. Oblak (University of Ljubljana, Ljubljana, Slovenia) Automatic Latent Fingerprint Segmentation Using Convolutional Neural Networks  Within the field of biometrics, fingerprint recognition is among the most popular approaches for automatic human identification. In order to successfully recognize individuals using fingerprints, the biometric samples need to be prepared first. This is particularly important for latent fingerprints, which are captured in unconstrained environments, e.g. crime scenes, and usually contain a significant amount of background noise. The first step is to segment the friction ridge impression from the background and determine the region of interest for further processing. In this paper we compare some of the most popular state-of-the-art methods for semantic image segmentation (FCN, UNet, SegNet, PSPNet) and use them to tackle latent fingerprint segmentation. We evaluate trained models on the IIITD MSLFD dataset and compare our results against the state-of-the-art for latent fingerprint segmentation, SegFinNet. We show that unmodified generic architectures are able to match the more complex SegFinNet in terms of false detection rate (FDR) and intersection over union (IoU), with greatly improved computational speed, reducing time required to process single batch by up to four times. 2. D. Cindori (Axiros, Zagreb, Croatia), I. Tomicic, P. Grd (Faculty of Organization and Informatics, Varaždin, Croatia) Security Hardening of Facial Recognition Systems  The usage of face recognition-based smart locks is growing daily in securing restricted physical locations, and although they provide an elegant and easily manageable access control solution, they might be subjected to various attacks such as with pre-recorded videos, 3D printed faces, and even with plainly printed static images of the authorized user. The main focus of this paper is to develop an additional security layer of a smart lock mechanism that would remedy such vulnerabilities and harden the facial recognition-based access control system. The implemented smart lock will use an IP camera near the protected door, and the facial recognition will be performed by an artificial intelligence algorithm running in the background of the Linux-based hardware. The smart lock system will be based on the affordable technology with limited processing power in mind. The security hardening will be based on the eye blink and liveness detection. Such hardened algorithm will prove to be more difficult to deceive, and at the actuator level will control the lock via relay mechanism in order to open the doors for authorized persons. 3. G. Oparin, V. Bogdanova , A. Pashinin (Institute for System Dynamics and Control Theory of SB RAS, Irkutsk, Russian Federation) Classification in Binary Feature Space Using Logical Dynamic Models  The article proposes a method based on the use of logical dynamic systems in the classification problem for Boolean vectors (binary feature vectors). This method has practical application in various fields of science and industry, in particular, socionics, geology, mechanical engineering, and others. Binary synchronous autonomous linear and nonlinear dynamic models with an unknown characteristic matrix are considered. Matrix elements are chosen the way that these binary feature vectors are equilibrium states of the binary dynamic model. The areas of attraction of equilibrium states act as classes. One reference vector corresponds to each class. The classification process consists of constructing trajectories of a logical-dynamic system with a given initial state. It aims to verify the reachability of one of the equilibrium states. A meaningful example of the proposed method is considered. 4. M. Štorga, M. Randić (Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Recognition of Symbols and Topology in the Image of a DC Circuit Diagram Based on Contours and Skeletons  Scope of research presented in this paper refers to the off-line recognition and interpretation of the content that conveys images of printed or hand-drawn direct current circuit diagrams. Photographing materials with different quality imprints as well as under different lighting exposures may introduce a lot of noise in the image. Besides the standard noise caused by shadows and unwanted spots, common noises that introduce ambiguities into the process of symbol and topology recognition are disconnections on lines and symbols. That is why we have designed algorithms in such a way to ensure correct recognition even in conditions of disconnections on lines and symbols. Due to the mostly regular geometric shapes and relationships that define symbols and the interconnection of symbols in an electrical circuit, and the clear topology related to the existence of nodes and branches, we decided to develop a recognition method based mostly on the structural relations. To recognize symbols and topology, we use suitable geometric primitives, contours, and skeletons that can be detected in the image. The recognition method based on the structural patterns that can handle disconnections represents the contribution of this paper. 5. A. Gribl, D. Petrinović (University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia) The Influence of Noise on 2D Gaussian Profile Parameters Estimation  The precise estimation of the Gaussian profile parameters plays an important role in many scientific fields. In astronomical images, the 2D Gaussian profiles are good approximations of stellar objects, which are point sources spread in the image plane due to various degradations in the imaging process. The accurate estimation of the 2D Gaussian parameters enables stellar object identification and outlier detection. However, various noise types present in astronomical images complicate the estimation process. Image noise can have an additive or multiplicative nature. The primary causes of additive Gaussian noise are thermal noise, which raises with the sensor temperature, and readout noise. The other type of noise is the multiplicative Poisson noise caused by the quantum nature of the light with the standard deviation proportional to the square root of the pixel intensity. Additionally, there are uniformly distributed quantization noise caused by the conversion of the continuous signal to discrete levels and fat-tailed distributed salt and pepper noise caused by the bad and hot pixels. All these noise types represent the challenge in the precise estimation of Gaussian parameters. Since the iteratively reweighted least squares method (IRWLS) with Huber weights calculated from statistical analysis of errors showed a good performance as a robust estimator in other applications, it is used in this paper for the 2D Gaussian profile parameters estimation from noisy data. The goal of this paper is to analyze the robustness of the IRWLS method to different noise types. 6. A. Skoki, J. Lerga, I. Štajduhar (Faculty of Engineering, University of Rijeka, Rijeka, Croatia) ML-Based Approach for NFL Defensive Pass Interference Prediction Using GPS Tracking Data  Defensive Pass Interference (DPI) is one of the most impactful penalties in the NFL. DPI is a spot foul, yielding an automatic first down to the team in possession. With such an influence on the game, referees have no room for a mistake. It is also a very rare event, which happens 1-2 times per 100 pass attempts. With technology improving and many IoT wearables being put on the athletes to collect valuable data, there is a solid ground for applying machine learning (ML) techniques to improve every aspect of the game. The work presented here is the first attempt in predicting DPI using player tracking GPS data. The data we used was collected by NFL’s Next Gen Stats throughout the 2018 regular season. We present ML models for highly imbalanced time-series binary classification: LSTM, GRU, ANN, and Multivariate LSTM-FCN. Results showed that using GPS tracking data to predict DPI has limited success. The best performing models had high recall with low precision which resulted in the classification of many false positive examples. Looking closely at the data confirmed that there is just not enough information to determine whether a foul was committed. This study might serve as a filter for multi-step pipeline for video sequence classification which could be able to solve this problem. 7. P. Bogović, A. Meštrović, S. Beliga, S. Martinčić-Ipšić (Sveučilište u Rijeci, Odjel za informatiku, Rijeka, Croatia) Topic Modelling of Croatian News During COVID-19 Pandemic  This paper addresses topic modelling in Croatian news articles related to COVID-19 pandemics and corresponding comments. We identify and analyze Croatian online news media's main topics for the first nine months of pandemics shedding some light on the leading themes covered in news articles and corresponding comments. Topics are derived automatically by training the Latent Dirichlet Allocation model and calculating topics' coherence values. We report the results by listing the top 15 detected words in top 10 detected topics from the content of articles and corresponding comments. Our findings include the analysis of intersected topics and discussion of dissents. Obtained results are the first step toward better information monitoring and hopefully mitigating the infodemics effect in Croatia. 8. P. Hernández García, D. Simunic (University of Zagreb, Zagreb, Croatia) Regulatory Framework of Artificial Intelligence in Healthcare  This paper provides an overview of the application of artificial intelligence in healthcare and what it means in many ways. These aspects will be the privacy that this new technology offers us versus the availability of information that this technology needs. We will also discuss the regulatory framework in the most important areas of the world such as the United States and Europe, comparing the laws and strategies that organizations have used to preserve the security and control of artificial intelligence inhealthcare. As a consequence, we will expose the ethical challenges posed by the entry of this new technology into our lives. We will also place ourselves in the current framework of the situation of artificial intelligence today, how it emerged, and its history over the years. To summarize, some conclusions have been proposed to conclude, and a personal opinion of authors is about everything discussed throughout the paperwork. 9. S. Delalić (Faculty of Science, Sarajevo, Bosnia and Herzegovina), E. Žunić (Faculty of Electrical Engineering, Sarajevo, Bosnia and Herzegovina), A. Alihodžić, E. Selmanović (Faculty of Science, Sarajevo, Bosnia and Herzegovina) A Discrete Bat Algorithm for the Rich Vehicle Routing Problem  The vehicle routing problem is one of the most complex problems in the field of combinatorial optimization. Creating optimal routes leads to timely delivery of orders to end customers, which increases the efficiency of the company and enables maximum earnings. The problem of vehicle routing with a series of real constraints is called the rich vehicle routing problem (RVRP). The paper presents an approach to solving RVRP, where the asymmetric routing problem with a heterogeneous vehicle fleet, time windows, customer-vehicle constraints and a number of others is observed. The approach solves the problem in two phases, by dividing customers into clusters using a discrete metaheuristic Bat algorithm, and by solving the routing problem for each obtained cluster. The proposed approach has been tested for more than 30 days of delivery from one of the largest warehouses in Bosnia and Herzegovina. Significant savings were achieved compared to previously implemented approaches. All created routes were feasible. The approach automatically creates routes, and gives results in a significantly shorter time than previous approaches. Time does not increase significantly with the increase in the number of customers. 10. A. Davydov, A. Larionov, N. Nagul (Matrosov Institute for System Dynamics and Control Theory SB RAS, Irkutsk, Russian Federation) Positively Constructed Formulas-Based Approach to Mobile Robot Control Using DES  The paper describes the principles of the implemen­tation of supervisors for discrete event systems (DES), designedwith the help of automatic theorem proving in the calculus ofpositively constructed formulas (PCFs). As an example, mobilerobot movement in an unknown environment is considered. Themain advantage of the proposed approach is the ability to carryout complex evaluations of events on the base of the current stateof the system and to use computable predicates, definitions ofwhich are based on the data coming from the external sources. 11. T. Bronzin, B. Prole, A. Stipić (CITUS, Zagreb, Croatia), K. Pap (University of Zagreb, Faculty of Graphic Arts , Zagreb, Croatia) Artificial Intelligence (AI) Brings Enhanced Personalized User Experience  The paper introduces new concepts of using artificial intelligence (AI) elements in culture and tourism industries by enhancing traditional user experiences with digital contents. This is done by creating a sophisticated backend system and integrating it with standard end-user devices like cell phones, tablets, and smart glasses. The proposed integration creates a new type of digital platform with wide range of possibilities. The platform allows automatic recognition of 2D images (like book pages, paintings, murals etc.) and adding the corresponding digital content (audio, video, text, links etc.) to significantly enhance user experience. The paper explains how this can be extended to include 3D functionality (at least under some restricted boundary conditions) without using expensive specialized 3D libraries or devices. Further integration with mapping and GPS services could also be used to prepare specialized tours, and even allow a self-service where end-user is allowed to create such tours from predefined elements. Such platform enables different items within the environment to trigger short messages (or even directly open digital contents) on the end-user’s device. 12. B. Ewerz (Virtual Vehicle Research GmbH, Graz, Austria), M. Hoefler (Karl-Franzen-Universität Graz, Graz, Austria), C. Marx, P. Moertl (Virtual Vehicle Research GmbH, Graz, Austria) How to Facilitate Explainability of AI for Increased User Trust: Results of a Study with a Covid-19 Risk Calculator   While the market of smart technologies is steadily increasing, there is much research to be done regarding the interaction between human users and Artificial Intelligence (AI) technologies. Specifically, the field of Explainable Artificial Intelligence (XAI) focuses on making AI explainable to users. To provide a user-centered approach to this growing field, this paper describes a study to investigate possible processes and methods. For this purpose, 20 participants were asked to use an AI system that provided them with the results of a personalized COVID-19 risk calculation. The study results indicate that while participants generally seemed to think that the presented results of the system were accurate, only a few said that they would change their behavior after receiving the results, and many asked for additional information to better understand the results. This paper discusses the findings along with possible approaches to increase behavior change in users of smart systems. 13. M. Jakovljević (Faculty of Traffic and Transportation Sciences, University of Zagreb, Zagreb, Croatia), S. Vojdović, P. Čolić (Ericsson Nikola Tesla, Zagreb, Croatia), O. Lale (Faculty of Traffic and Transportation Sciences, University of Zagreb, Zagreb, Croatia), E. Jovandžikov (Ericsson Nikola Tesla, Zagreb, Croatia) Methodology for Assessment of Mobile Telecom Databased Origin-Destination Matrices Accuracy  Origin Destination Matrices (ODM) are standard indicators of population migration used for urban mobility assessment. They represent a population’s travel demand (number of trips between two zones) and are considered as the step two of a four-step model of transportation planning process. The research community has already recognized that ODM obtained from different sources might differ and have proposed several assessment methods in order to determine the difference among them. Besides a literature review, this research will provide a methodology for the analysis of similarity of ODMs generated from telecom big data sets compared to those obtained using other measurements. To test and validate the proposed methodology, ODM has been created using telecom big data analytics, where a significant number of anonymized telecom user locations have been analyzed to determine a travel demand. The travel demand characteristic for the same area and the same period of time has in parallel been determined/measured using “the traditional” methodology resulting in the second data set. The paper will provide the analysis results of the comparison of two resulting matrices, lessons learnt, and recommendations that can significantly improve the result correlation rate. Finally, the research has proved that a significant correlation between the resulting matrixes can be achieved if the appropriate preconditions are met. 14. K. Vidović (Ericsson Nikola Tesla, Zagreb, Croatia), M. Šoštarić (Faculty of Traffic and Transportation Sciences, University of Zagreb, Zagreb, Croatia), A. Blavicki (Ericsson Nikola Tesla, Zagreb, Croatia), F. Sirovica (Faculty of Traffic and Transportation Sciences, University of Zagreb, Zagreb, Croatia) Validation Points in Process of Urban Mobility Assessment Using Telecom Big Data Analytics  Big data sets originating from mobile telecommunication networks, besides their primary purpose within the telecom environment, are becoming more popular in other application areas and in other verticals. Number of researches have been performed where the analytics based on those data sets have proven applicable in use cases related to urban mobility. Besides positive results of those researches, urban mobility researches, planners, and stakeholders in general still heavily rely and base their decisions (urban mobility policies) on traditional, engraved transportation related researches/measurements (e.g. traffic counting, surveying, census data). This article proposes novel approach for urban mobility assessment that utilizes both telecoms originated big data analytics, as primary data source for urban mobility assessment, and traditional measurements as secondary data source. Traditional measurements performed within this research are used for validation of the big data analysis. This article defines and elaborates on number of validation points within the process of urban mobility assessment, where the results of both big data and traditional methodology (used as ground truth) are compared and validated. 15. L. Jovova, K. Trivodaliev (Faculty of Computer Science and Engineering, Ss. Cyril and Methodius University, Skopje, Skopje, Macedonia) Air Pollution Forecasting Using CNN-LSTM Deep Learning Model  One of the greatest issues modern urban environments are facing is poor air quality. It directly affects human health having a long-term negative impact on people's lives and is a major cause of deaths in the world. Smart cities combined with advances in deep learning provide a novel platform for dealing with this problem. This paper uses pollution data from smart sensor networks and a CNN-LSTM architecture to forecast the air pollution concentration of the current hour based on the previous 24-hour pollution concentration and several meteorological features from the previous hour. Initially data is preprocessed with special focus and strategy for handling missing values. The performance of the model is fine-tuned by taking into account additional temporal and seasonal dependency of this type of data. Comparison with other models from classical machine learning shows that the proposed deep learning model has better performance according to the provided metrics. 16. M. Vitek, A. Hafner, P. Peer, A. Jaklič (Faculty of Computer and Information Science, University of Ljubljana, Ljubljana, Slovenia) Evaluation of Deep Approaches to Sclera Segmentation  Sclera segmentation is an important processing step in a biometric system based on this modality. We propose adaptations of five deep neural network architectures: SegNet, DeepLabv3+, HRNetV2, UPerNet, and U-Net for sclera segmentation. The architectures are experimentally evaluated on recent MASD and SBVPI eye image datasets using standard metrics: precision, recall, $F_{1}$-score and intersection over union. Experiments on the SBVPI dataset show that the performance for sclera segmentation of the five deep neural network architectures is very similar in terms of the used metrics and acceptable for practical applications. Additionally, the SegNet, HRNetV2 and UPerNet, and U-Net tested on MASD perform better in the terms of precision metrics than the winning U-Net adaptation from the Sclera Segmentation Benchmark Competition 2019 (SSBC 2019), with our U-Net being the best deep model. 17. E. Stefanovska, A. Kralevska, A. Abazi, A. Naumoski (Faculty of computer science and engineering, Ss. Cyril and Methodius University in Skopje, Skopje, Macedonia) A GIS Based Approach for Finding Association between Air Pollution, Factories and Population Density in North Macedonia  Air pollution is a serious problem in North Macedonia. In this project all the air quality stations in North Macedonia were mapped along with the closest factories to these stations. Later, the collected data was used to analyze how the number of factories and the population density around air quality stations correlates with the amount of pollution measured by given station. Comparison between air quality data collected during whole year and only in winter months was also made. This whole process was carried in ArcGIS, a software which has variety of spatial visualization and processing tools. Our work shows that, as expected, there is a positive correlation between the number of factories close to a given station and the pollution that it measures. The same pattern is true for the population density, too. We hope that this research paper will have a positive impact for solving the air pollution problem and perhaps changes towards improving the whole situation will be made, especially in the areas identified with the highest amount of pollution. 18. E. Simikj, F. Bozinovski, A. Naumoski (Faculty of computer science and engineering, Ss. Cyril and Methodius University in Skopje, Skopje, Macedonia) Hospitals and Pharmacy Relationship between Rating, Service Time and Closeness in the City of Skopje  Due to the urgent global health situation, the hospitals as well as the pharmacy stores are under pressure. Therefore, the aim of our project is to use the geo-spatial information of these two objects in Skopje, the capital of North Macedonia, and to represent their detail information to find the closest pharmacy near the hospital, rate the hospitals and pharmacies based on user customers experience, and provide this information to customers to make online ordering and pick up the order later. To do this, we have collected the data about the hospitals and pharmacy objects, then we import this data in ArcGIS software, and we produce distance, interpolation, and service maps from the data analysis. As results, the final visual representation from our analysis shows possible improvements and possible new locations for pharmacy stores. 19. N. Kale, P. Grd, I. Tomičić (Fakultet organizacije i informatike, Varaždin, Croatia) Automatska usporedba tetovaža  Posljednjih godina u području biometrije sve se više istraživanja fokusira na meke biometrijske karakteristike. Meke karakteristike daju određene informacije o osobi, a mogu biti fizičke ili ponašajne. Primjeri mekih biometrijskih karakteristika su boja kose, boju očiju, dob, spol, etnička pripadnost, boja kože, visina, težina, ožiljci i tetovaže. Analizom mekih biometrijskih karakteristika možemo doći do mnogih informacija o osobi, a jedna od takvih karakteristika su i tetovaže. Primjena analize tetovaža je naročito česta u policiji i pravosuđu kao pomoć u verifikaciji osumnjičenika ili žrtvi u istragama. Cilj ovog članka je dati pregled dosadašnjih istraživanja na temu automatske detekcije i prepoznavanja tetovaža te prikazati implementaciju metode za usporedbu tetovaža. Predložena metoda ima nekoliko koraka: predprocesiranje, izlučivanje karakteristike i usporedba. U koraku predprocesiranja slika se pretvara u sliku u nijansama sive te se koristi Canny detektor rubova kako bi se detektirali rubovi tetovaže. U sljedećem koraku izlučuju se karakteristike tetovaže iz rubova detektiranih u prethodnom koraku korištenjem SIFT (Scale-Invariant Feature Transform) algoritma. Zadnji korak je usporedba izlučenih karakteristika i računanje podudaranja metodom najbližih susjeda.

Slobodan Ribarić (Croatia), Andrea Budin (Croatia)

Patrizio Campisi (Italy), Bojan Cukic (United States), Ivo Ipšić (Croatia), Marina Ivašić-Kos (Croatia), Ruizhe Ma (United States), Neeta Nain (India), Nikola Pavešić (Slovenia), Vitomir Štruc (Slovenia), Zheng-Hua Tan (Denmark)

The discount doesn't apply to PhD students.

Slobodan Ribaric
University of Zagreb
Faculty of Electrical Engineering and Computing
Unska 3
HR-10000 Zagreb, Croatia

Phone: +385 1 612 99 52
Fax: +385 1 612 96 53
E-mail: slobodan.ribaric@fer.hr

 
Andrea Budin
Ericsson Nikola Tesla Inc.
Krapinska 45
HR-10000 Zagreb, Croatia

Phone:+385 1 365 34 23
Fax: +385 1 365 3548
E-mail: andrea.budin@ericsson.com

The best papers will get a special award.
Accepted papers will be published in the ISSN registered conference proceedings. Presented papers in English will be submitted for inclusion in the IEEE Xplore Digital Library.
.............
There is a possibility that the selected scientific papers with some further modification and refinement are being published in the Journal of Computing and Information Technology (CIT).