Cracks are the most requiring type of pavement distresses to detect and classify automatically. Due its nature easily absorbed by other types surface damages. Moreover, diversity makes image detection system efficient computer algorithms. The paper presents solutions tested on distress data which were collected automatically using downward facing cameras placed orthogonally road axis. Presented results focus crack-type distresses. achieved accuracy transverse, longitudinal meshing cracks...

The magnetorheological elastomer is promising material for applications in soft robotics. Its properties like reactive to external magnetic field and softness allow construct an attractive devices. This work presents a construction of gripper assembled with elastomers. describes the detailed molding process Further, electromechanical elastomers are shown using simple beam. Finally, constructed analyzed series experiments.

This paper presents a novel approach to recognize scene presented in an image with specific application classification field sports video. We propose different variants of the algorithm ranging from bags visual words simplified real-time implementation, that takes only most important areas similar colour into account. All feature accuracy which is comparable very well-known indexing techniques like SIFT or HoGs. For comparison purposes, we also developed database now available online. The...

This paper presents hardware implementation of most commonly used MPEG-7 color descriptors. The testing circuits was described using VHDL language and synthesized into FPGA. proposed system architectures are for description in real-time image's basing on features such as: distribution, spatial layout structure color. split the computational burden several processes where calculations mentioned made simultaneously order to improve system's speed. These methods make realizations main...

The paper presents hardware implementation of the MPEG-7 edge histogram descriptor. testing circuit was described using VHDL language and synthesized into FPGA. RC1000 board with a Xilinx Virtex V1000 FPGA chosen as target platform. Experimental results descriptor efficiency are presented too

Many image processing operations such as: segmentation, contour tracking, morphological operations, etc. needs an additional information about pixel neighborhood. In the other hand memory access are often (comparing to working time of hardware devices) time-consuming, so checking values neighboring pixels via could lead low performance algorithm. Our binary representation contains it's neighborhood on one byte only, which is stored at pixel's address. Moreover, because zeros in pixels,...

Paper presents a novel idea of failure detection mechanism for complex control environments. The is composed several dynamic artificial neural networks that work in parallel order to detect failing signal from one the on-board robot sensors. simulation results show system capable detecting quickly and efficiently.

In this paper we focus on the stochastic kinetic extension of well-known Hodgkin-Huxley model a biological neuron. We show gradient descent algorithm for training neuron model. comparison with use only three weights instead nine. that trained gives equally good results as model, while gain more concise mathematical description procedure. The is tested in solving problem approximation, where approximated function membrane potential obtained using different models was chosen. Additionally,...

This paper presents a deep learning-based image texture recognition system.The methodology taken in this solution is formed bottom-up manner.It means we swipe moving window through the order to categorize if given region belongs one of classes seen training process.This categorization done based on Deep Neural Network (DNN) fixed architecture.The process fully automated regarding data preparation, investigation best algorithm, and its hyper-parameters.The only human input system definition...

Convolutional Neural Networks (CNNs) have become a standard approach to many image processing dilemmas. Consequently, most of the proposed CNN architectures tend increase model deepness or layer complexity. Thus, they are composed parameters and need considerable computing resources training examples. However, some recent works show that either shallow neural networks without convolutions can achieve similar results with these models often being used in systems limited resources....

This work presents the architecture of communication interface between nodes dedicated cluster (custom computing machine) for electromagnetic field computations by finite elements method. The channel is based on LVDS (low voltage differential signaling) standard. Basics modules transmitter/receiver units are also discussed. In addition paper project testing board, which will be used validation tests proposed channel.

The paper presents a novel approach to image preprocessing for feature extraction that is designed reduction of dimensionality the classifier which in this case convolutional neural network (CNN). proposed method uses Restricted Boltzmann Machine(RBM) as an Aggregation Method (AM) binary descriptors. assumption technique RBM performing dimension expansion space. Also type data undergoes transformation from floating point. conventional networks input consists one (grayscale) or three channels...

This paper describes the approach to collision avoidance problem for 3-DOF anthropomorphic robot manipulators. The novelty of is decomposition 3D space two 2D spaces. Resulting computationally efficient algorithm, suitable implementation in real-time systems. Simulation manipulator operating three dimensional with obstacles also presented.

Road pavement structure analysis provides important information for road maintenance and renovation planning. Image processing methods applying to cross-section of the asphalt can provide valuable about its quality. The paper presents a method analyze picture sample, taking into account aggregates size. features that have been captured quantified using 2-D images were limited elaborated algorithm implementation are presented. tool is prepared as part bigger system dedicated analysis.

This study addresses the problem of collision-free controlling 3-DOF (degree freedom) anthropomorphic manipulators with given a priori unrestricted trajectory. The robot constraints resulting from physical robot's actuators are also taken into account during movement. Obstacle avoidance algorithm is based on penalty function, which minimized when collision predicted. Mathematical construction function and minimization process allows modeling variety behaviors elusion moves. Implementation...