János Hollósi

57188990738

Publications - 18

Driver Distraction Detection in Extreme Conditions Using Kolmogorov–Arnold Networks

Szabolcs Fischer János Hollósi Mykola Sysyn Gábor Kovács Dmytro Kurhan Viktor Nagy

Publication Name: Computers

Publication Date: 2025-05-01

Volume: 14

Issue: 5

Page Range: Unknown

Description:

Driver distraction can have severe safety consequences, particularly in public transportation. This paper presents a novel approach for detecting bus driver actions, such as mobile phone usage and interactions with passengers, using Kolmogorov–Arnold networks (KANs). The adversarial FGSM attack method was applied to assess the robustness of KANs in extreme driving conditions, like adverse weather, high-traffic situations, and bad visibility conditions. In this research, a custom dataset was used in collaboration with a partner company in the field of public transportation. This allows the efficiency of Kolmogorov–Arnold network solutions to be verified using real data. The results suggest that KANs can enhance driver distraction detection under challenging conditions, with improved resilience against adversarial attacks, particularly in low-complexity networks.

Open Access: Yes

DOI: 10.3390/computers14050184

Detection of Bus Driver Mobile Phone Usage Using Kolmogorov-Arnold Networks

Á Ballagi Szabolcs Fischer János Hollósi Gábor Kovács Viktor Nagy

Publication Name: Computers

Publication Date: 2024-09-01

Volume: 13

Issue: 9

Page Range: Unknown

Description:

This research introduces a new approach for detecting mobile phone use by drivers, exploiting the capabilities of Kolmogorov-Arnold Networks (KAN) to improve road safety and comply with regulations prohibiting phone use while driving. To address the lack of available data for this specific task, a unique dataset was constructed consisting of images of bus drivers in two scenarios: driving without phone interaction and driving while on a phone call. This dataset provides the basis for the current research. Different KAN-based networks were developed for custom action recognition tailored to the nuanced task of identifying drivers holding phones. The system’s performance was evaluated against convolutional neural network-based solutions, and differences in accuracy and robustness were observed. The aim was to propose an appropriate solution for professional Driver Monitoring Systems (DMS) in research and development and to investigate the efficiency of KAN solutions for this specific sub-task. The implications of this work extend beyond enforcement, providing a foundational technology for automating monitoring and improving safety protocols in the commercial and public transport sectors. In conclusion, this study demonstrates the efficacy of KAN network layers in neural network designs for driver monitoring applications.

Open Access: Yes

DOI: 10.3390/computers13090218

Bus Driver Head Position Detection Using Capsule Networks under Dynamic Driving Conditions

Á Ballagi Szabolcs Fischer János Hollósi Gábor Kovács Viktor Nagy

Publication Name: Computers

Publication Date: 2024-03-01

Volume: 13

Issue: 3

Page Range: Unknown

Description:

Monitoring bus driver behavior and posture in urban public transport’s dynamic and unpredictable environment requires robust real-time analytics systems. Traditional camera-based systems that use computer vision techniques for facial recognition are foundational. However, they often struggle with real-world challenges such as sudden driver movements, active driver–passenger interactions, variations in lighting, and physical obstructions. Our investigation covers four different neural network architectures, including two variations of convolutional neural networks (CNNs) that form the comparative baseline. The capsule network (CapsNet) developed by our team has been shown to be superior in terms of efficiency and speed in facial recognition tasks compared to traditional models. It offers a new approach for rapidly and accurately detecting a driver’s head position within the wide-angled view of the bus driver’s cabin. This research demonstrates the potential of CapsNets in driver head and face detection and lays the foundation for integrating CapsNet-based solutions into real-time monitoring systems to enhance public transportation safety protocols.

Open Access: Yes

DOI: 10.3390/computers13030066

Efficiency Analysis of Kolmogorov-Arnold Networks for Visual Data Processing †

János Hollósi

Publication Name: Engineering Proceedings

Publication Date: 2024-01-01

Volume: 79

Issue: 1

Page Range: Unknown

Description:

In the field of artificial neural networks, the use of multilayer perceptrons (MLPs) has long been a well-established methodology. Recently, the theory of Kolmogorov–Arnold Networks (KANs) has emerged as a potential alternative to multilayer perceptrons, inspired by the Kolmogorov–Arnold representation theorem. It has been demonstrated that solutions based on the Kolmogorov–Arnold Network (KAN) can achieve better efficiency than those based on the multilayer perceptron (MLP) for certain problems. In this work, we investigate how the new theory can be applied to a special image classification task when some adversarial attack method is applied. The aim of the research is to explore the potential of the theory to answer the question of its applicability to complex tasks of practical importance.

Open Access: Yes

DOI: 10.3390/engproc2024079068

Face Detection Using a Capsule Network for Driver Monitoring Application

Á Ballagi Szabolcs Fischer János Hollósi Gábor Kovács Viktor Nagy

Publication Name: Computers

Publication Date: 2023-08-01

Volume: 12

Issue: 8

Page Range: Unknown

Description:

Bus driver distraction and cognitive load lead to higher accident risk. Driver distraction sources and complex physical and psychological effects must be recognized and analyzed in real-world driving conditions to reduce risk and enhance overall road safety. The implementation of a camera-based system utilizing computer vision for face recognition emerges as a highly viable and effective driver monitoring approach applicable in public transport. Reliable, accurate, and unnoticeable software solutions need to be developed to reach the appropriate robustness of the system. The reliability of data recording depends mainly on external factors, such as vibration, camera lens contamination, lighting conditions, and other optical performance degradations. The current study introduces Capsule Networks (CapsNets) for image processing and face detection tasks. The authors’ goal is to create a fast and accurate system compared to state-of-the-art Neural Network (NN) algorithms. Based on the seven tests completed, the authors’ solution outperformed the other networks in terms of performance degradation in six out of seven cases. The results show that the applied capsule-based solution performs well, and the degradation in efficiency is noticeably smaller than for the presented convolutional neural networks when adversarial attack methods are used. From an application standpoint, ensuring the security and effectiveness of an image-based driver monitoring system relies heavily on the mitigation of disruptive occurrences, commonly referred to as “image distractions,” which represent attacks on the neural network.

Open Access: Yes

DOI: 10.3390/computers12080161

Simplified Routing Mechanism for Capsule Networks

R. C. Pozna Á Ballagi János Hollósi

Publication Name: Algorithms

Publication Date: 2023-07-01

Volume: 16

Issue: 7

Page Range: Unknown

Description:

Classifying digital images using neural networks is one of the most fundamental tasks within the field of artificial intelligence. For a long time, convolutional neural networks have proven to be the most efficient solution for processing visual data, such as classification, detection, or segmentation. The efficient operation of convolutional neural networks requires the use of data augmentation and a high number of feature maps to embed object transformations. Especially for large datasets, this approach is not very efficient. In 2017, Geoffrey Hinton and his research team introduced the theory of capsule networks. Capsule networks offer a solution to the problems of convolutional neural networks. In this approach, sufficient efficiency can be achieved without large-scale data augmentation. However, the training time for Hinton’s capsule network is much longer than for convolutional neural networks. We have examined the capsule networks and propose a modification in the routing mechanism to speed up the algorithm. This could reduce the training time of capsule networks by almost half in some cases. Moreover, our solution achieves performance improvements in the field of image classification.

Open Access: Yes

DOI: 10.3390/a16070336

Capsule-based Autoencoder Network for Pointcloud Reconstruction

R. C. Pozna Á Ballagi János Hollósi

Publication Name: 2023 IEEE 21st World Symposium on Applied Machine Intelligence and Informatics Sami 2023 Proceedings

Publication Date: 2023-01-01

Volume: Unknown

Issue: Unknown

Page Range: 121-126

Description:

The theory of capsule networks and the dynamic routing mechanism for capsules was introduced by Geoffrey Hinton and his research team. In this new approach, they tried to solve typical problems of classical convolutional neural networks. For example, that the efficiency of neural networks degrades when a geometric transformation is applied on the input image, or when the data is far away from the training dataset. It became clear early on that capsule networks are state-of-the-art solutions for visual data classification tasks. For other tasks their use is less common and in many cases difficult to apply. For example image segmentation or object detection and localization. The efficiency of the capsule networks theory in the field of pointcloud processing is also an open question. In this work we investigated the pointcloud reconstruction capability of capsule networks. In this approach, three different complexity autoencoder networks was selected. We created a decoder network based on capsules theory, which was fitted to the existing autoencoder networks. The efficiency of the networks was tested using four different datasets. As a result of our work, we show the effectiveness of capsule networks in the field of pointcloud reconstruction compared with the selected autoencoder networks.

Open Access: Yes

DOI: 10.1109/SAMI58000.2023.10044532

Capsule Network based 3D Object Orientation Estimation

R. C. Pozna Á Ballagi János Hollósi

Publication Name: International Conference on Electrical Computer Communications and Mechatronics Engineering Iceccme 2023

Publication Date: 2023-01-01

Volume: Unknown

Issue: Unknown

Page Range: Unknown

Description:

Convolutional neural networks have proven to be one of the most efficient methods for processing visual data. Due to the popularity of the field, there is a growing interest in the reliability of intelligent systems. It has been shown that convolutional neural networks can be fooled by extreme inputs or noisy inputs. To overcome the current problems of convolutional neural networks, the theory of capsule networks was introduced by Geoffrey Hinton and his research team. In this work we want to investigate the theory of capsule networks for orientation recognition of 3-dimensional objects. We consider the case when the data are noise loaded by various adversarial attacking methods. We compare our results with the efficiency of convolutional neural network based solutions, highlighting the difference between the two theories. We investigate the efficiency reduction that can be observed using different adversarial attacking methods. Our results will show how much more efficient the capsule network is compared to the neural networks.

Open Access: Yes

DOI: 10.1109/ICECCME57830.2023.10252762

Digital Twin of Drone-based Protection of Agricultural Areas

Á Ballagi Csaba Hajdu A. J. Kovács János Hollósi Gergely Teschner Norbert Boros

Publication Name: 2022 IEEE 1st International Conference on Internet of Digital Reality Iod 2022

Publication Date: 2022-01-01

Volume: Unknown

Issue: Unknown

Page Range: 99-104

Description:

Protecting agricultural fields, like crops, vineyards, and husbandry areas, has been a difficult challenge since historical times. Classical methods to prevent intrusion are often destructive to wild and domestic animals alike. Even more current nondestructive systems, like camera-based systems are attributed to specific problems related to environmental or technological issues. Furthermore, verifying the effectiveness of installed systems is difficult, as the triggering situations are unmanageable and typically occur unsupervised. This paper presents a complex vision-based intrusion detection system to overcome these problems and further proposes more extensive control and flexibility on the development process. The solution provides a workflow integrating Digital Reality methods into the system development by creating a digital twin of the drone and its surrounding environment in a general-purpose robotic simulator. With this simulation, the triggering events and environmental effects can be easily emulated, such as a wild animal entering the area of interest. The solution also focuses on incorporating new 5G info-communication networks on handling communication between the intrusion detection system and the base station in a distributed manner.

Open Access: Yes

DOI: 10.1109/IoD55468.2022.9986763

Economical Mobile Robot Design Prototype and Simulation for Industry 4.0 Applications

R. C. Pozna Á Ballagi Csaba Hajdu János Hollósi Rudolf Krecht

Publication Name: Cando EPE 2020 Proceedings IEEE 3rd International Conference and Workshop in Obuda on Electrical and Power Engineering

Publication Date: 2020-11-18

Volume: Unknown

Issue: Unknown

Page Range: 155-160

Description:

Autonomous mobile robots received a rising research interest, with the appearance of complex cyber-physical system applications, like Industry 4.0. Related scenarios require not only some degree of autonomous capabilities of the control software but also tight interaction between humans as well. In this paper, a prototype autonomous mobile robot of such functionalities is presented with additionally aiming cost-effective construction. The robot is currently capable of performing basic tasks, like traversing to a goal configuration, mapping the environment, and detecting and avoiding obstacles. To allow the versatile extension with additional cyber-physical software, the control software interfaces the Robot Operating System (ROS), a popular framework in robotic research. During the development of software and hardware, simulation has been heavily utilized. This allowed an iterative development and the use of verification in the early development phases.

Open Access: Yes

DOI: 10.1109/CANDO-EPE51100.2020.9337786

Training Neural Networks with Computer Generated Images

Á Ballagi János Hollósi

Publication Name: Informatics 2019 IEEE 15th International Scientific Conference on Informatics Proceedings

Publication Date: 2019-11-01

Volume: Unknown

Issue: Unknown

Page Range: 155-159

Description:

At the Széchenyi István University we develop an autonomous racing car for the Shell Eco-marathon. One of the main tasks is to create a neural network which is segment the road surface, the protective barriers and other components of the race track. The difficulty of this task, that there is no a right dataset for this special issue. Only a limited size dataset available, therefore, we would like to expands this dataset with computer generated training images, which comes from a virtual city environment. In this work we want to examine the effect of computer generated images on the efficiency of different neural networks. In the training process real images and computer generated virtual images are mixed in several different ways. After that, three different neural network architecture for road surface and road barrier detection are trained. Experiences shows how to mixing datasets and how they can improve efficiency.

Open Access: Yes

DOI: 10.1109/Informatics47936.2019.9119273

Training Capsule Networks with Various Parameters

Á Ballagi János Hollósi

Publication Name: Saci 2019 IEEE 13th International Symposium on Applied Computational Intelligence and Informatics Proceedings

Publication Date: 2019-05-01

Volume: Unknown

Issue: Unknown

Page Range: 191-196

Description:

Nowadays convolutional neural networks (CNNs) have produced the state-of-The-Art performance in image classification and segmentation tasks. The efficiency of the neural networks is one of the bests when the testing samples are close to the training data. Nevertheless, if we make some transformation on the dataset, the performance of the convolutional neural network reduced. Recently, capsule networks (CapsNet) have been introduced to solve some of the problems of neural networks. In this paper we examine the effectiveness of three different capsule based neural networks, and compare the performance when the parameters of the dynamic routing algorithm and the squashing function are modified.

Open Access: Yes

DOI: 10.1109/SACI46893.2019.9111574

Demonstrating the Options for Automated Advanced Selective Waste Gathering

István Lakatos Adam Titrik János Hollósi

Publication Name: Sisy 2018 IEEE 16th International Symposium on Intelligent Systems and Informatics Proceedings

Publication Date: 2018-11-05

Volume: Unknown

Issue: Unknown

Page Range: 167-171

Description:

Due to continuous developments, the exhausting and unpleasant work done by humans is going to be partly or totally replaced by automated devices. Such developments can be found in the field of transport concerning autonomous and networked vehicles. Although we are at a quite early stage of development, it is highly practical to make plans for a more distant future, i.e. targeting a totally automated selective waste gathering. Apart from the application of smart containers, this study presents alternatives for networked automated selective waste gathering.

Open Access: Yes

DOI: 10.1109/SISY.2018.8524707

Improve the Accuracy of Neural Networks using Capsule Layers

R. C. Pozna János Hollósi

Publication Name: 18th IEEE International Symposium on Computational Intelligence and Informatics Cinti 2018 Proceedings

Publication Date: 2018-11-01

Volume: Unknown

Issue: Unknown

Page Range: 15-18

Description:

Neural networks are a powerful and widely used tools for various classification and segmentation tasks. Nowadays, in the field of computer vision the convolutional neural networks (CNNs) are the most popular solution for many problems. The CNNs performance is looks like exceptionally great when the test images are very close to the training dataset. However if the input images are transformed, such as rotating or tilting, the efficiency of the neural network may be greatly reduced. A new kind of neural network, is called capsule network, is trying to solve this problem. In this paper we examines the efficiency of the capsule network, by trying to increase the accuracy of different networks with capsule layers.

Open Access: Yes

DOI: 10.1109/CINTI.2018.8928194

Two-stage racetrack segmentation method using color feature filtering and superpixel-based convolutional neural network

R. C. Pozna Ernö Horváth János Hollósi

Publication Name: Saci 2018 IEEE 12th International Symposium on Applied Computational Intelligence and Informatics Proceedings

Publication Date: 2018-08-20

Volume: Unknown

Issue: Unknown

Page Range: 131-135

Description:

The Széchenyi István University race car team is an active and successful participant of the Shell Eco-marathon for long time ago. The Shell introduces the autonomous vehicle category on the Eco-marathon for 2018. Our long-term goal is to make the Szenergy racing team's vehicle suitable for the autonomous category. The first milestone is to make a reliable computer vision based intelligent detection system that understands the environment of the racing car. In this paper we will present a solution for racetrack detection i.e. a fusion of image processing and neural network systems. The two-stage recognition system is at the first phase an image processing algorithm which finds the red-white and blue-white striped edge of the road, and at the second phase, a pre-trained superpixel-based neural network which recognize the road on the filtered image.

Open Access: Yes

DOI: 10.1109/SACI.2018.8440968

A use case of the simulation-based approach to mobile robot algorithm development

R. C. Pozna Ernö Horváth János Hollósi Csaba Hajdu

Publication Name: Sami 2016 IEEE 14th International Symposium on Applied Machine Intelligence and Informatics Proceedings

Publication Date: 2016-03-01

Volume: Unknown

Issue: Unknown

Page Range: 311-314

Description:

A complex algorithm development progress for mobile robot (MR) usually requires a priori simulation of the whole dynamic environment before it is applied on a real-world robot. In order to reduce the development time and avoid programming mistakes it is recommended to use the same language for both simulation and real-word testing. The paper presents a certain use case for the simulation-based approach. In this use case V-REP is used for simulation, ROS for the real-world robot application and MATLAB or C# for algorithm development. Note that the source code of the work is available on a public GitHub repository.

Open Access: Yes

DOI: 10.1109/SAMI.2016.7423026

The inverse kinematics problem, a heuristical approach

R. C. Pozna Ernö Horváth János Hollósi

Publication Name: Sami 2016 IEEE 14th International Symposium on Applied Machine Intelligence and Informatics Proceedings

Publication Date: 2016-03-01

Volume: Unknown

Issue: Unknown

Page Range: 299-304

Description:

This paper presents a heuristic solution for the inverse kinematics problem. The heuristic consists on combining the distance between the actual position and the desired position of the gripper with the direction of the best manipulability of the robot. Theoretical results are validated by digital simulations resolute.

Open Access: Yes

DOI: 10.1109/SAMI.2016.7423024

YOLO-Based Object and Keypoint Detection for Autonomous Traffic Cone Placement and Retrieval for Industrial Robots

János Hollósi

Publication Name: Applied Sciences Switzerland

Publication Date: 2025-10-01

Volume: 15

Issue: 19

Page Range: Unknown

Description:

The accurate and efficient placement of traffic cones is a critical safety and logistical requirement in diverse industrial environments. This study introduces a novel dataset specifically designed for the near-overhead detection of traffic cones, containing both bounding box annotations and apex keypoints. Leveraging this dataset, we systematically evaluated whether classical object detection methods or keypoint-based detection methods are more effective for the task of cone apex localization. Several state-of-the-art YOLO-based architectures (YOLOv8, YOLOv11, YOLOv12) were trained and tested under identical conditions. The comparative experiments showed that both approaches can achieve high accuracy, but they differ in their trade-offs between robustness, computational cost, and suitability for real-time embedded deployment. These findings highlight the importance of dataset design for specialized viewpoints and confirm that lightweight YOLO models are particularly well-suited for resource-constrained robotic platforms. The key contributions of this work are the introduction of a new annotated dataset for overhead cone detection and a systematic comparison of object detection and keypoint detection paradigms for apex localization in real-world robotic applications.

Open Access: Yes

DOI: 10.3390/app151910845