This research article presents the utilization of an adaptive camera system for virtual teleport in videoconferencing, enabling users to experience remote locations in an immersive and interactive manner. The camera system incorporates features such as motion of camera system in X and Y axes, object tracking, and detection using Ultralytics YOLO V5 (You Only Look Once). The live video feed from the camera system is streamed over the internet by using OBS (Open Broadcaster Software version 29.1.3) virtual camera via platforms like Google Meet and viewed through VR headsets. Users can control the camera's movement and direction using intuitive interfaces like joysticks, enhancing their sense of presence and immersion. This article highlights the design and implementation of the adaptive camera system, its integration with virtual reality technologies, and the user experience it offers. The findings demonstrate the potential of this system for providing a realistic and engaging virtual teleportation experience in future.
In this paper we explore the possibilities of using Neural Networks with transfer learning approach and Dynamic Time Warping for online signature verification. Verification process of proposed methods is based on SVC-2004 signature database, which contains genuine and skilled forgery signatures. Experimental results show that approach using Convolutional Neural Networks is more successful and this method is achieving an Equal Error Rate (EER) as low as 2.25% using 10-fold cross-validation. These results are competitive with other published works.
As part of the ‘Zero Emission Sail Ship’ project, one of the research tasks was to develop an automatic sailing system. To achieve this, we created a mathematical model of the steering system for the three-mast schooner Klara, which has the same hull as the future zero-emission sailing ship. The paper describes the identification procedure for both linear and nonlinear steering ship models, based on several sea trials with the sailing ship Klara. The best of these models will be used in automatic sailing system simulation.
A new CMOS active resistor structure is submitted. This new proposed symmetrical structure offers the possibility to obtain a strong improvement of the main circuit performances, as compared to the classical similar designs. This circuit current-voltage characteristic is very linear and the domain of the differential input voltage allowed by the circuit is very large, for obtaining this excellent linearity. As a result of the biasing in saturation of all MOS active devices, frequency response of the new proposed structure will be strongly increased. This original circuit allows a very important diminution of the circuit area as compared to physical resistors, accurately simulating the Ohm law for the equivalent resistances up to megaohms. The SPICE simulations validate the theoretical approach, showing a very good linearity, a large domain of the input voltage also a very good frequency response.
Detecting brain aneurysms is a critical area of research due to the potential life-threatening consequences of aneurysm rupture. Early detection of a cerebral (brain) aneurysm can increase the chances of successful treatment. In this paper, we explore various neural network architectures such as 2DCNN, PointCNN, PointNet and 3DCNN for classification of cerebral aneurysms. Our main objective is to develop a reliable system that enhances the accuracy of aneurysm detection by medical professionals. Ultimately, the goal is to contribute towards improving patient outcomes and saving lives. Our experimental results, which were obtained using the IntrA: 3D Intracranial Aneurysm dataset, indicate the superiority of the proposed 2DCNN and 3DCNN for detection of brain aneurysms. We compare the results with other deep learning architectures. Our proposed 2DCNN and 3DCNN architectures achieved precision of 85.91% and 89.05%, respectively.
Insulator burn mark detection is a critical component of Micro Air Vehicle (MAV) inspection, which has evolved into the mainstream of high-voltage power line inspection. This paper addressed the challenge of burn mark inspection on power line defects in real-time, given a low volume of the collected dataset and under less human supervision. We introduced two methods for the intelligent real-time inspection of burn marks based on passive and active approaches. The proposed Models were based on YOLOv5 and achieved a frame rate of 50 FPS. The test results indicated that the proposed active deep learning approach could effectively perform insulator inspection and accurately identify several types of burn mark defects. Our proposed approach can achieve a high accuracy of mAP 0.989% while reducing the annotation overhead to less than 57% and reducing the computational cost. Moreover, the proposed active inspection model actively seeks feedback from both human annotators and other sources to tune the burn mark detection. As a result, the model can continue to learn from its errors and enhance its precision over time. In addition, it can deliver better generalization and interpretability by actively seeking diverse and representative samples during the training process.
The goal of the research is to compare the built-in renderers in two software, one free and the other licensed, Blender and Autodesk 3Ds Max by comparing their renders of the interior model. In the experimental part models of furniture and other interior elements were made. The comparison of two 3D models, bathroom and living room, with different lighting conditions, natural and artificial light source was done visually.
This article proposes a method for simple and reliable encryption of digital images using chaotic transformations of Arnold's cat and Baker maps. The transformed source image is then combined using the XOR operation with a random image generated by a simple sensor of a pseudo-random sequence of numbers distributed according to a uniform distribution law. The study conducted by the authors shows the impossibility of statistical attacks on an encrypted image by intruders due to its high cryptographic strength.
Software-defined networks are slowly but surely coming into production. Currently, there are several different SDN architectures. As SDN solutions continue to evolve, there is room to improve current solutions in networks with multiple SDN controllers. One of the researched aspects of the SDN network is the deployment of multiple controllers to improve reliability, load balancing, and efficiency. Each controller will manage its SDN domain. In this paper, we proposed an efficient SDN architecture based on a flat multicontroller multidomain network that provides efficient load balancing in the network. We demonstrated and verified its performance via simulations in Mininet environment with Ryu controllers.
Many of the renewable energy generation methods suffer from the variable power output they produce. This is because they depend on elements with a substantial stochastic nature. To mitigate the variability different controlling mechanisms must be deployed. For thorough and optimal control, it is a necessity to predict the generated power in the near or more distant future. In this article, we explore the possibility of predicting electricity production by photovoltaic panels based on sky images. In addition, we also tested selected meteorological data that are relevant and available. For this purpose, we used SkyCam dataset that contains required information. We have proposed 3 scenarios combining block-based and sequential processing of sky images and selected meteorological data using CNN and LSTM networks. The best results were obtained by a combination of sky images and meteorological data processed as sequence of frames by LSTM. The achieved mean absolute error was 44.3 W/m 2 for 15-minute intervals.
Human activities, primarily through greenhouse gas emissions, have accelerated global warming. As the transportation sector accounts for roughly one-quarter of all greenhouse gas emissions, an imperative is to find efficient transport solutions that reduce our footprint. Advancing this goal, in this paper, we leverage state-of-the-art techniques to compute optimal energy and travel time paths on the road network, thus enabling conventional and electric vehicle users to traverse the roads efficiently by reducing time spent in congestion and promoting routes with lower energy consumption. Consequently, significantly reducing their emissions footprint. The results present the opportunity and effectiveness of employing graph theory algorithms to address the identified issues and minimise travel time and energy consumption.
This study explored the application of artificial neural networks in predicting age of the neonatal brain based on the structural magnetic resonance images (MRI). We implemented regression, a classical method for predicting continuous values using supervised learning, based on the fastMONAI library. We tested the performance of the implemented neural network on the T 1 and T 2 weigthed neonatal brain MRI from the publicly available dHCP (Developing Human Connectome Project) dataset. Research was conducted on data subsets consisting of 10, 100, and 200 T 1 and T 2 weigthed images. Additionally, we explored the effect of using a constant learning rate versus automatically finding the optimal learning rate. Finally, we performed training, validation, and testing of the algorithm with the optimal parameters. The obtained results on the test data subset achieved an error of less than 1.5 weeks of age.
This paper deals with examining the possibility of peer grading when ranking graphical user interface (GUI) design images. The assessment was made on real images created by students on a GUI design course. Students' assignments to design a graphical user interface were being subjected to pairwise comparison and ranking was made based on pairwise comparison matrix. Obtained students' ranking was compared to experts' opinion using Spearman's and Kendall's metrics. The results suggest the possibility of using students' peer to peer grades to obtain objective rank from their subjective comparisons.
The aim of the paper is the quality of video streaming analysis in cases of using different video codecs in the environment of distributed computer systems with different QoS (Quality of Service). For the purposes of the analysis, several scenarios were set up in which video encoded with different codecs is transmitted by a virtual video streaming server to virtual clients. For each of the scenarios, an environment with different QoS (packet losses, latency, jitter) was simulated and the quality of the received video stream was evaluated for each video codec. The quality of the received decoded video stream was calculated using SSIM (Structural Similarity) and VMAF (Video Multimethod Assessment Fusion) video objective metrics and compared to the original video stream.
Video processing tasks require the analysis of an enormous amount of data. However, in most practical applications this analysis is not required on the whole frame or video but is limited to the regions where some action is taking place. If these regions can be identified with low effort, the total computation time required to process and analyze such videos can be drastically reduced. In this work a new motion saliency detection algorithm intended to be incorporated in a human action recognition pipeline is presented. The approach uses depth data to improve the performance of the difference of frames method which is computationally simple and efficient. Results show that the algorithm achieves a performance which is comparable to the state of the art while requiring a much lower computational time.
There are two major challenges in using electronic circuits that mimic negative capacitors and negative inductors (non-Foster elements): ensuring stable operation and minimizing the dependence of the generated negative immittance on frequency. Recently, it has been shown that there is an inevitable trade-off between these two goals, caused by the background physics of used negative impedance converters (NICs). In order to provide guidelines for practical design, we analyze the dispersion curves for open-circuit stable and short-circuit stable non-Foster elements based on NICs with one-pole amplifiers and develop associated equivalent circuits.
Recently, a tunable bandpass filter with active impedance inverter comprising non-Foster negative capacitor based on well-known Linvill's design, has been proposed. Here, we extend this idea by using a non-Foster negative capacitor that comprises compensated passive RL structure. It was found that the new filter had properties similar to the one based on originally proposed approach but with better stability and sensitivity properties.
Fire presents a dangerous occurrence in inhabited places or places occupied by humans. Apart from very high material damage, it can result in human fatalities. In the past, numerous sensor-based methods were developed in order to reduce the reaction time and improve fire detection accuracy. With the advent of modern convolutional neural networks (CNN), object detection models quickly emerged, together with novel fire detection methods utilizing said models. In this paper, we aim to demonstrate the performance of YOLOv5 on our custom dataset, which contains indoor fire occurrences. Furthermore, we are investigating the influence of the input image resolution on the models' performances. This is important due to the hardware limitations of the models that would be used in real-life applications. The research shows that different metrics (F1-Score, inference time, mAP50) can yield different models as the best-performing ones. However, since the models have to be relatively small, quick, and accurate, we have proposed the ranking-based evaluation of the models from the aspect of the input image resolution. The evaluation showed that the models obtained the best overall score when the input image resolution was set to $512 \times 512$ pixels.
One of the most common causes of road accidents is the driver's inattention or the underestimation of the current situation. This problem can be amplified on the less maintained road sections, where the critical parts (curves) are not visible. These factors often result in fatalities or other serious injuries. With the increasing popularity of connected and automated vehicles, Vehicle-to-Everything (V2X) communication has emerged as a promising technology to prevent head-on collisions in critical road sections. In this paper, we propose a head-on collision avoidance system that utilizes current vehicle information on the monitored road section to detect potential collisions and provide sufficient warnings to drivers. This system is designed to cooperate with multiple other systems to create a comprehensive solution for potential autonomous vehicles. The entire solution was also extensively tested using custom road scenarios and simulated vehicular data using a high-fidelity traffic simulation generated by SUMO. The resulting application could be used and extended for future uses for head-on collision avoidance or the implementation of autonomous vehicles.
This paper provides insight into the migration of smart metering and advanced metering infrastructure into the cloud and using cloud computing as a means for companies to provide the services without the need to possess their own infrastructure. Firstly, we pointed out the main issues of the conventional system and explained the general idea of applying cloud computing for smart metering and advanced metering infrastructure. Secondly, we discussed the different software, platform, and infrastructure models that can be used depending on the type of services being provided. Next, we explained the different types of deployments that the cloud could be categorized into. After this, we listed and commented on some benefits and drawbacks of cloud computing for the application of smart metering and advanced metering infrastructure. Finally, we showcased the forerunners of cloud service providers.
