Pavel Škrabánek

Labyrinth Arrangement Analysis Based on Image Processing

The approach described in this paper has been developed as a supporting tool for practical exercises based on a teaching aid; however, its application is far from being limited to be used only in the education process. The main task of the approach is extraction of information about structure of a working environment from color images and its transformation to an appropriate form suitable for path-planning. The working environment considered in this paper is a labyrinth. Since the information is to be used further by path-planning, its expression in the form of a graph is required, or, more precisely, the adjacency matrix is required as the output. The transformation of the real world into a discreet representation is based on the exact cell decomposition in this approach. The approach employs the well-known image processing algorithms; however, the procedure of the labyrinth layout analysis as well as the transformation of the acquired information into the adjacency matrix has been developed and tested in this work. The approach has been realized and verified in MATLAB using the Image Processing Toolbox.

The time optimal path-planning of mobile robots motion respecting the time cost of rotation

The paper describes the time optimal path-planning method designed for differential wheeled mobile robots operating on flat ground. The robots are used as support teaching tool by the path-planning problematic exercising. Whereas the exercise is designed for students without any prior knowledge about the path-planning, the graph version of the A* algorithm was chosen as the appropriate algorithm for the problematic introduction. The students are supposed to exercise the path-planning using the evaluation functions of various difficulties. The most complex of them is the evaluation function reflecting both the transportation time and the time required for a robot rotation. Its mathematical formulation is described in the paper and its functionality is shown in three case studies where the shortest time-path between two locations in a labyrinth is required to be found.

Magnetic strips as landmarks for mobile robot navigation

In this paper, a low-cost localization system designed for mobile robots is introduced. The system has been developed for a teaching aid where a mobile robot operates in a maze. The mobile robot is navigated by a high level control system where path-planning is executed; however, autonomous execution of the plan is required from the robot. The robot uses magnetic landmarks, among others, for its localization. Detection of the landmarks is realized by a triple axis magnetometer sensor. The landmarks are made from a commercially available magnetic sheet. Construction of the sheets limits this approach. A successful application has to meet some requirements which are described in this paper.

Detection of grapes in natural environment using feedforward neural network as a classifier

The recognition of wine grapes in images acquired in natural environment is a serious issue solved by researches dealing with precision viticulture. The detection of wine grapes of red kinds is a well managed problem. On the other hand, the detection of white grapes is still a challenging task. In this contribution, the classifier for white wine grapes recognition is introduced and evaluated. The classifier is based on an artificial neural network and is used in two ways which differ in image representation. Namely, the pixel intensities and histogram of oriented gradients are used for the representation of images. Then, feedforward multilayer neural network is applied as a classifier. The classifiers based on the histograms of oriented gradients seemed to be very effective - they were almost error free from the cross validation point of view and they performed well with the independent testing data sets, too. On the other hand, the representation using pixel intensities was stated as insufficient for classification using our approach.

Extraction of outliers from imbalanced sets

In this paper, we presented an outlier detection method, designed for small datasets, such as datasets in animal group behaviour research. The method was aimed at detection of global outliers in unlabelled datasets where inliers form one predominant cluster and the outliers are at distances from the centre of the cluster. Simultaneously, the number of inliers was much higher than the number of outliers. The extraction of exceptional observations (EEO) method was based on the Mahalanobis distance with one tuning parameter. We proposed a visualization method, which allows expert estimation of the tuning parameter value. The method was tested and evaluated on 44 datasets. Excellent results, fully comparable with other methods, were obtained on datasets satisfying the method requirements. For large datasets, the higher computational requirement of this method might be prohibitive. This drawback can be partially suppressed with an alternative distance measure. We proposed to use Euclidean distance in combination with standard deviation normalization as a reliable alternative.

Reflecting on Imbalance Data Issue When Teaching Performance Measures

Importance of soft computing methods has continuously grown for many years. Particularly machine learning methods have been paid considerable attention in the business sphere and subsequently within the general public in the last decade. Machine learning and its implementation is the object of interest of many commercial subjects, whether they are small companies or large corporations. Consequently, well-educated experts in the area of machine learning are highly sought after on the job market. Most of the technical universities around the world have incorporated the machine learning into their curricula. However, machine learning is a dynamically evolving area and the curricula should be continuously updated. This paper is intended to support this process. Namely, an imbalance data issue, in context of performance measures for binary classification, is opened, and a teaching method covering this problem is presented. The method has been primary designed for undergraduate and graduate students of technical fields; however, it can be easily adopted in curricula of other fields of study, e.g. medicine, economics, or social sciences.

High-Level Control System for a Path-Planning Teaching Aid

The paper brings a description of a high-level control system which is a part of a teaching aid aimed at practicing path-planning methods. The teaching aid uses a proven concept of a mobile robot operating within a maze. The high-level control system ensures path-planning, data collection, data processing and data distribution. This contribution covers topics related to the development of a software part of the high-level control system. Specifically, software requirements, software design, and software testing are detailed in the text.

Robust Grape Detector Based on SVMs and HOG Features

Detection of grapes in real-life images is a serious task solved by researchers dealing with precision viticulture. In the case of white wine varieties, grape detectors based on SVMs classifiers, in combination with a HOG descriptor, have proven to be very efficient. Simplified versions of the detectors seem to be the best solution for practical applications. They offer the best known performance versus time-complexity ratio. As our research showed, a conversion of RGB images to grayscale format, which is implemented at an image preprocessing level, is ideal means for further improvement of performance of the detectors. In order to enhance the ratio, we explored relevance of the conversion in a context of a detector potential sensitivity to a rotation of berries. For this purpose, we proposed a modification of the conversion, and we designed an appropriate method for a tuning of such modified detectors. To evaluate the effect of the new parameter space on their performance, we developed a specialized visualization method. In order to provide accurate results, we formed new datasets for both tuning and evaluation of the detectors. Our effort resulted in a robust grape detector which is less sensitive to image distortion.

Pattern recognition neural network as a tool for pest birds detection

Various kinds of vermin have been considered as a huge problem since primeval times. Over this period, means of protection against vermin have developed to be very quick and efficient. However, new goals in protection have appeared recently which reflects legislative changes in most countries. Public opinion has shifted towards greater environment protection. Nowadays, vermin control systems have turned from being used globally into local applications and from being applied preventively into casual usage. Thus, accurate vermin detection units are becoming very important parts of vermin control systems. This situation is valid in agricultural areas (e.g. vineyards) which are protected against pest birds, too. Reflecting on the current situation, a feedforward multilayer artificial neural network, aimed on detection of European starling in vineyards, is presented in this paper. Except a description and validation of the detection method, the idea of the comprehensive protection system is also outlined in this paper.

Neural Network as a Tool for Detection of Wine Grapes

The recognition of wine grapes in real-life images is a serious issue solved by researches dealing with precision viticulture. The detection of wine grapes of red varieties is a well mastered problem. On the other hand, the detection of white varieties is still a challenging task. In this contribution, detectors designed for recognition of white wine grapes in real-life images are introduced and evaluated. Two representations of object images are considered in this paper; namely, vector of normalized pixel intensities and histograms of oriented gradients. In both cases, classifiers are realized using feedforward multilayer neural networks. The detector based on the histograms of oriented gradients has proven to be very effective by cross-validation. The results obtained by its evaluation on independent testing data are slightly worse; however, still very good. On the other hand, the representation using the vector of normalized pixel intensities was stated as insufficient.