Search

This paper presents a new approach to processing and categorization of text from patient documents in Bulgarian language using Natural Language Processing and Edge AI. The proposed algorithm contains several phases - personal data anonymization, pre-processing and conversion of text to vectors, model training and recognition. The experimental results in terms of achieved accuracy are comparable with modern approaches.

The GNSS system allows on a planetary scale to fulfill navigation functions, that is, to determine the current location of consumers of navigation information and their speed and to carry out accurate time on the Earth's surface and in near-Earth space. There are several sources of errors that distort the code and the carrier of the navigation signal. The noise floor of the receiver error and the errors from multipath propagation of the signal are independent for each receiver. Multipath propagation of the signal is the main source of errors for high-precision navigation. In this paper we propose a methods for estimation of these errors, which are based on the properties of the time correlation of the multipath propagation of the navigation signal.

In this paper we present techniques based on machine learning techniques on monitoring data for analysis of cybersecurity threats in cloud environments that incorporate enterprise applications from the fields of telecommunications and IoT. Cybersecurity is a term describing techniques for protecting computers, telecommunications equipment, applications, environments and data. In modern networks enormous volume of generated traffic can be observed. We propose several techniques such as Support Vector Machines, Neural networks and Deep Neural Networks in combination for analysis of monitoring data. An approach for combining classifier results based on performance weights is proposed. The proposed approach delivers promising results comparable to existing algorithms and is suitable for enterprise grade security applications.

In this article is presented an overview of the existing solutions for smart grids, and an analysis of the majority cases of cybersecurity threats. The purpose of this article is to define the risks of compromising the confidentiality of the transferred and stored data, and the ability to install a new device into the network. The main hazards coming from unauthorized access to the smart grid devices are analyzed and systematized and recommendations are given to improve security. The proposed smart grid design aims at making the system as safe as possible from the viewpoint of information protection and communication using intelligent communication and control devices.

Performance estimation of large scale cloud deployments of web servers is an ongoing challenge for researchers and operations engineers all over the globe. In this paper we propose an approach for performance estimation of cloud environment based on performance metrics and scalable coefficient between number of backend nodes and number of load balancers. An example performance estimation using this methodology is provided for a large automated environment in a public cloud.

In this article we propose a design for infrastructure to service a new type of e-Learning environment for technology studies. The idea of the authors is to lay the foundations of an eLearning platform which is efficient, scalable, supportable and with reduced total cost of ownership. The proposed components are frontend portal for teachers and students hosted on an web server; backend servicing the portal with load-balancing component and open-source relational database. The various resources is provided by relatively affordable hardware and open-source software such as KVM, Docker and Puppet.

This paper presents an approach for automated estimation of human emotions from electroencephalogram data. The used features are principally the Hjorth parameters calculated for theta, alpha, beta and gamma bands taken from certain channels. The classification stage is support vector machine. Since the human emotions are modelled as combinations from physiological elements such as arousal, valence, dominance, liking, etc., these quantities are the classifier's outputs. The best achieved correct classification performance is about 80%.

Capturing the type of physical activity a person is performing thorough his daily life, can inspire the development of new and innovative applications. Examples include monitoring patients' health and physical activity performance, reasoning upon the observed activity to recommend better training strategy, new therapeutic programs, etc. In this work we propose an algorithm for Human Activity Recognition based on the application of a geometrically motivated feature selection method. We test the algorithm on a standard data set and validate its performance by comparing it with the existing results of other known algorithms.

In this paper a system is presented for classification of facial images based on facial age estimation. The presented age-group classification system can be used for multimedia forensics where the investigation can be facilitated by automated analysis of large image datasets. Hence, two main features of such systems are usually desired: increased speed of computation and reliability to real-life images. In an attempt to meet these criteria we propose a combination of subspace projection algorihtm and a classifier for age group classification. Additionally a face detection algorithm is integrated to ensure detection of faces in complex scenes. The proposed system is implemented as software application and was tested on a large image dataset with real-life capturing conditions.

This paper considers the problem of face detection and tracking with an active camera. An algorithm using Haar-like face detector and Mean-shift method for face tracking has been presented. We propose an adaptive algorithm that provides automated control of a Pan-Tilt camera (PTC) to follow a person’s face and keep their image centered in the camera view. An error vector defined in the image plane and representing the face offset with respect to the center of the image is used for camera control. Adaptive feedback control design and stability analysis are performed via Lyapunov techniques. Simulation results are presented to illustrate the effectiveness of the proposed algorithms.

In this paper an approach for personal biometric identification is presented based on extraction of ECG features and classification with RBFNN. We perform denoising and segmentation on the input signal, after which we realize dimensionality reduction and feature extraction based on PCA transform. The separability of the selected features is improved by applying LDA. The final stage of the proposed approach is classification and recognition of the extracted features with classifier score fusion.

The paper describes combined approach to face detection for grayscale images using the combined cascade of neural network classifiers which consists of Haar-like features' cascade of weak classifiers and convolutional neural network. The combined cascade with proposed face candidates' verification method allows achieving one of the best detection rates on CMU test set and a high processing speed suitable for a video flow processing. In the last stage of our algorithm we provide extraction of salient facial features for pose estimation and recognition.

In this article an algorithm is proposed for estimation of eye-pupil contour parameters in an image, by means of a particle filter and artificial infrared lighting. A description is given of the initial assumptions that must be made when the algorithm is used. Then the used system model is described, as well as the principles of evolving the state of the system in time. Also the method is depicted, for estimation of the hypotheses of eye-pupil presence in the analyzed image. This is necessary for the application of particle filtering in the algorithm. In addition, it is necessary to introduce a means for reduction of the number of hypotheses. We have also shown and analyzed the experimental results from the proposed algorithm.

This paper presents an approach to face detection and tracking in video sequences. We use a set of Haar-like features to train a cascade of classifiers. Face-tracking is performed using the mean-shift algorithm. We propose a method for adaptation both of the tracking-window and of the color-distribution model in order to increase robustness to illumination changes of the environment. Minimizing of the number of iterations is achieved by using dynamic prediction with Kalman filter.

In this paper we present a novel approach to recognition of faces in frontal color images. It involves face extraction, creation of face models with wavelet packet decomposition for dimensionality reduction, Principal Component Analysis (PCA) of the decomposed faces, Linear Discriminant Analysis (LDA) over the PCA subspace, neural classifiers with radial basis functions for each modality and combination of classifier results. The first step of the method is face detection through skin-color modeling and segmentation. After the face is extracted, wavelet decomposition is performed. PCA follows the wavelet-packet decomposition, after which further clustering in the subspace is performed using LDA. Then, neural classifiers are created respectively for the wavelet coefficients, PCA projections, and LDA projections. In the end, combination of classifier results is used to improve overall system availability. We tested our approach on a single database. The proposed approach delivered excellent results.