Vol. 36 · No. 2

This paper describes a non-intrusive real-time gaze detection system, characterized by a precise determination of a subject's pupil centre. A narrow field-of-view camera (NFV), focused on one of the subject's eyes follows the head movements in order to keep the pupil centred in the image. When a tracking error is observed, feedback provided by a second camera, in this case a wide field-of-view (WFV) camera, allows quick recovery of the tracking process. Illumination is provided by four infrared LED blocks synchronised with the electronic shutter of the eye camera. The characteristic shape of corneal glints produced by these illuminators allows optimizing the image processing algorithms for gaze detection developed for this system. The illumination power used in this system has been limited to well below maximum recommended levels. After an initial calibration procedure, the line of gaze is determined starting from the vector defined by the pupil centre and a valid glint. The glints are validated using the iris outline to avoid glint distortion produced by changes in the curvature on the ocular globe. In order to minimize measurement error in the pupil-glint vector, algorithms are proposed to determine the pupil centre at sub-pixel resolution. Although the paper describes a desk-mounted prototype, the final implementation is to be installed on board of a conventional car as an embedded system to determine the line of gaze of the driver.

Several algorithms have been proposed in the last years for discovering stops in trajectories of moving objects. Some methods consider as stops the subtrajectories that i) have speed lower than the average trajectory speed, ii) present significant direction changes, iii) have gaps, or iv) intersect a given spatial region. In these approaches a time constraint should be met for the subtrajectory to be considered as a stop, and this constraint is absolute (it is met or not). Indeed, these approaches consider stops as a continuous subtrajectory. In this paper, we show that for several application domains the stops do not need to be continuous, and the time constraint should be relaxed. In summary, we present the definitions of non-continuous stops and present an algorithm to discover a new kind of stops. We evaluate the proposed algorithm with a running example and real trajectory data, comparing it to the most similar approach in the literature, the SMoT algorithm.

In this paper, we propose new real-time measurement-based scheduling algorithms to achieve a trade-off between the energy efficiency and the performance capability of computational clusters. An investigation is performed using a specific scenario of computing clusters with realistic parameters. Numerical results show that a trade-off between the performance and the energy efficiency can be controlled by the proposed algorithms.

The article presents a metaheuristic solution for the problem of shape optimization of a rotating annular disc. Such discs are important structural components of e.g. jet engines, steam turbines or disc brakes. The design goal is to find the disc shape that would ensure its maximal carrying capacity (corresponding to the speed of rotation), which is a variational problem with the objective functional defined by L infty norm. Such a definition makes the problem impossible to solve using analytical methods so utilization of metaheuristics is necessary. We present different algorithms to solve the problem starting with a classic evolutionary one, followed by agent-based and hybrid agent-based memetic algorithms, which are the main focus of this paper. The reason for this is that agent-based computing systems proved to be versatile as an optimization technique being especially efficient for the problems with complex fitness functions. The obtained experimental results encourage further application of such an approach to similar engineering problems.

Developing peer-to-peer (P2P) applications became increasingly important in software development. Nowadays, a large number of organizations from many different sectors and sizes depend more and more on collaboration between actors to perform their tasks. These P2P applications usually have a recursive behavior that many modeling approaches cannot describe and analyze (e.g. finite-state approaches). In this paper, we present an approach that combines component-based development with well-understood methods and techniques from the field of Attribute Grammars and Data-Flow Analysis in order to construct an abstract representation (i.e. Data-Dependency Graph) for P2P applications, and then perform data-flow analyzes on it. This approach embodies a formalism called DDF (Data-Dependency Formalism) to capture the behavior of P2P applications and construct their Data-Dependency Graphs. Various properties can be inferred and computed at the proposed level of data abstraction, including some properties that model checking cannot compute if the system presents a recursive behavior. As examples, we present two algorithms: one to resolve the deadlock problem and another for dominance analysis.

Architecture-centric development is one of the most promising methods for improving software quality, reducing software cost and raising software productivity. Software architecture research not only focuses on the design phase, but also covers every phase of software life cycle. Software architecture has characteristics different from traditional software, conventional testing methods do not apply directly to software architecture. Basis path testing is a very simple and efficient white-box testing method. Traditional methods generate basis path according to the control flow graph, they are not suitable for generating component path when we detect more software architecture errors. This paper presents a new concept – Basis Component Path (BCP) for C2-style architecture, and proposes a method to generate the BCPs. C2-style architecture is represented by components, connectors, and interfaces, and uses an architecture component interaction graph (CIG) to describe interface connection relationship. We also provide an algorithm to generate BCP set. Experiments apply the proposed method in a typical C2-style architecture and the result shows that the proposed method can generate BCP set which contains as many BCPs as possible efficiently, and it meets the requirements of the basis component path testing.

In recent years, real-time multimedia applications' demands such as Voice-on-IP (VoIP) and video conference are extremely increased which require QoS routing. This type of routing has been considered as an NP-Complete problem since it requires satisfying multiple constraints. Many solutions have been proposed to solve it, but most of them are complex and time consuming. In this paper, a novel multi-constraints QoS routing algorithm is proposed based on Evolutionary Strategies (ES). The algorithm preserves simplicity and offers a feasible solution in a few numbers of generations. This is due to a novel gene decoding mechanism that is used in the algorithm; and consequently more simple evolutionary operators can be applied. The simulation results show that our method outperforms previous algorithms in terms of speed and performance, so that it is 2.6 and 11.3 times faster, and its success ratio is also better.

This paper presents a parallel implementation of the fast isogeometric solvers for explicit dynamics for solving non-stationary time-dependent problems. The algorithm is described in pseudo-code. We present theoretical estimates of the computational and communication complexities for a single time step of the parallel algorithm. The computational complexity is O(p6 N/c t comp) and communication complexity is O(N/(c(2/3)t comm) where p denotes the polynomial order of B-spline basis with Cp-1 global continuity, N denotes the number of elements and c is number of processors forming a cube, t comp refers to the execution time of a single operation, and t comm refers to the time of sending a single datum. We compare theoretical estimates with numerical experiments performed on the LONESTAR Linux cluster from Texas Advanced Computing Center, using 1 000 processors. We apply the method to solve nonlinear flows in highly heterogeneous porous media.

In this paper, we propose a three-parametric convolution kernel which is based on the one-parameter Keys kernel. The first part of the paper describes the structure of the three-parameter convolution kernel. Then, a certain analytical expression for finding the position of the maximum of the reconstructed function is given. The second part presents an algorithm for estimating the fundamental frequency of the speech signal processing in the frequency domain using Picking Picks methods and parametric cubic convolution. Furthermore, the results of experiments give the estimated fundamental frequency of speech and sinusoidal signals in order to select the optimal values of the parameters of the proposed convolution kernel. The results of the fundamental frequency estimation according to the mean square error are given by tables and graphics. Consequently, it is used as a basis for a comparative analysis. The analysis derived the optimal parameters of the kernel and the window function that generates the least MSE. Results showed a higher efficiency in comparison to two or three-parameter convolution kernel.

In the big data era, learning-based techniques have attracted more and more attention in many industry areas such as smart grid, intelligent transportation. The power load forecasting is one of the most critical issues in data analysis of smart grid. However, learning-based methods have not been widely used due to the poor data quality and computational capacity. In this paper, we propose a comprehensive learning-based model to forecast heavy and over load (HOL) accidents according to the data from various information systems. At first, we present a combined random under- and over-sampling technique for imbalanced electric data, and choose an optimal sampling rate through several experiments. Then, we reduce the attributes that have significant impact on the power load by using learning-based methods. Finally, we provide an algorithm based on the random forest method to prevent the over-fitting problem. We evaluate the proposed model and algorithms with the real-world data provided by China Grid. The experimental results show that our model works efficiently and achieves low error rates.