Object recognition and classification as well as obstacle distance calculation are of the utmost importance in today’s autonomous driving systems. One such system designed to detect obstacle and track intrusion in railways is considered in this paper. The heart of this system is the decision support system (DSS), which is in charge of making complex decisions, important for a safe and efficient autonomous train drive based on the information obtained from various sensors. DSS determines the object class and its distance from a running train by analyzing sensor images using machine learning algorithms. For the quality training of these machine learning models, it is necessary to provide training sets with images of adequate quality, which is often not the case in real-world railway applications. Furthermore, the images of insufficient quality should not be processed at all in order to save computational time. One of the most common types of distortion which occurs in real-world conditions (train movement and vibrations, movement of other objects, bad weather conditions, and day and night image differences) is blur. This paper presents an improved edge-detection method for the automatic detection and rejection of images of inadequate quality regarding the blur level. The proposed method, with its improvements convenient for railway application, is compared with several other state-of-the-art methods for blur detection, and its superior overall performance is demonstrated.

This paper presents a new method for designing adaptive neuro-fuzzy inference systems (ANFIS). Improvements are made by introducing specially developed orthogonal functions into the very structure of ANFIS, specifically, into the layer that imitates Sugeno stile defuzzification. These functions are specially tailored for analysis and synthesis of dynamic systems and they also contain an adaptive measure of the variability of the systems operating in a real environment, which can be implemented inside the ANFIS as hormonal effect.

In this article, a new class of quasi-orthogonal filters, based on the Legendre and Malmquist-type quasi-orthogonal polynomials, is presented. These filters are generators of quasi-orthogonal functions for which we derive and present all important properties and relations. Our article is based on the classical theory of orthogonality and orthogonal functions, and also on new results in this field of mathematics. Based on theoretical results, we design schemes for the realisation of these filters. Finally, a trail quasi-orthogonal filter is practically realised and its quasi-orthogonality is proven by performing experiments. Quasi-orthogonal filters can be successfully used for signal approximation as well as for modelling, identification, analysis, synthesis and simulation of dynamical systems.

This paper presents a new approach for improving performances of magnetic levitation system. Controlled parameter is the amplitude which levitated object achieves during movement from one levitation position to another. Two position levitation with improved amplitude performances is obtained by implementing modified neural network in standard magnetic levitation control logic. The modified network is a nonlinear autoregressive neural network with newly developed activation function based on orthogonal polynomials. Performed experiments on a system with default control logic showed that it could not provide stable two position levitation when specified amplitude of the levitation object is greater than 10^-4m. Artificial network was trained using real experimental data and it was based on standard tangent and sigmoid activation functions. Default activation functions were then substituted with a newly developed orthogonal polynomial function. The amplitude 10^-3m was achieved with stable two position levitation after parameter optimization. It is proven that simple control logic with nonlinear autoregressive neural network and proper activation function can provide improved amplitude performances. DOI: http://dx.doi.org/10.5755/j01.eie.23.3.13167

In this article, a new intelligent hybrid controller is proposed. The controller is based on the combination of the orthogonal endocrine neural network (OENN) and orthogonal endocrine ANFIS (OEANFIS). The orthogonal part of the controller consists of Chebyshev orthogonal functions, which are used because of their recursive property, computational simplicity, and accuracy in nonlinear approximations. Artificial endocrine influence on the controller is achieved by introducing excitatory and inhibitory glands to the OENN part of the structure, in the form of postsynaptic potentials. These potentials provide a network with the capability of additional self-regulation in the presence of external disturbances. The intelligent structure is trained using a developed learning algorithm, which consists of both offline and online learning procedures: online learning for fitting OENN substructure and offline learning for adjusting OEANFIS parameters. The learning process is expanded by introducing the learning rate adaptation algorithm, which bases its calculations on the sign of the error difference. Finally, the proposed intelligent controller was experimentally tested for control of a nonlinear multiple-input-multiple-output two rotor aerodynamical system. During the test phase, an additional four related intelligent control logics and default PID-based controllers were used, and tracking performance comparisons were performed. The proposed controller showed notably better online results in comparison to other control algorithms. The major deficiencies of the structure are complexity and noticeably large training computation time, but these drawbacks can be neglected if tracking performances of a dynamical system are of the highest importance.