Streszczenie: Systemy wizyjne to dynamicznie rozwijająca

Due to technological progress in forestry, seedlings with covered root systemsespecially those grown in container nurserieshave become increasingly important in forest nursery production. One the trees that is most commonly grown this way is the common oak (Quercus robur L.). For an acorn to be sown in a container, it is necessary to remove its upper part during mechanical scarification, and evaluate its sowing suitability. At present, this is mainly done manually and by visual assessment. The low effectiveness of this method of acorn preparation has encouraged a search for unconventional solutions. One of them is the use of an automated device that consists of a computer vision-based module. For economic reasons related to the cost of growing seedlings in container nurseries, it is beneficial to minimize the contribution of unhealthy seeds. The maximum accuracy, which is understood as the number of correct seed diagnoses relative to the total number of seeds being assessed, was adopted as a criterion for choosing a separation threshold. According to the method proposed, the intensity and red components of the images of scarified acorns facilitated the best results in terms of the materials examined during the experiment. On average, a 10% inaccuracy of separation was observed. A secondary outcome of the presented research is an evaluation of the ergonomic parameters of the user interface that is attached to the unit controlling the device when it is running in its autonomous operation mode.

The paper presents an application for generating ground truth data for the purposes of video detection and justifies their use in systems which analyze road traffic videos. The usefulness of described application in the development of video detection software is presented especially during scene configuration and comparative analysis of video detection results versus ground truth data. The latter is possible due to simplicity of the result text files generated in a similar way both by the presented application and by the video detection algorithm. Two example applications of the tool designed to generate ground truth data are presented, together with a discussion of their construction, functionality and abilities. ZBIGNIEW BUBLINSKI, PIOTR PAWLIK AGH University of Science and Technology, Faculty of Electrical Engineering, Automatics, Computer Science and Biomedical Engineering, Department of Automatics and Biomedical Engineering {buba,piotrus}@agh.edu.pl SEGMENTATION OF OBJECTS IN LOG-POLAR SPACE Abstract. The aim of this work was to develop a simple and fast yet robust algorithm for segmentation of objects in log-polar space. Assuming that some interesting parts of an image are selected by fixation of point of attention, pixels located close to fixation points are transformed to a log-polar space, then edge detection is performed. As a final step, detected edges are processed in a specific manner, which should lead to obtain boundaries of objects. After some smoothing, the detected boundaries are transformed back to a cartesian space and are superimposed on an original image in order to show results of segmentation. The aim of this work was to develop a simple and fast yet robust algorithm for segmentation of objects in log-polar space. Assuming that some interesting parts of an image are selected by fixation of point of attention, pixels located close to fixation points are transformed to a log-polar space, then edge detection is performed. As a final step, detected edges are processed in a specific manner, which should lead to obtain boundaries of objects. After some smoothing, the detected boundaries are transformed back to a cartesian space and are superimposed on an original image in order to show results of segmentation. Joanna Grabska-Chrząstowska *, Przemysław Korohoda **, Zbigniew Pieprzak*** * AGH University of Science and Technology, Faculty of Electrical Engineering, Automatics, Computer Science and Biomedical Engineering, Department of Automatics and Bioengineering, Al. Mickiewicza 30, 30-059 Krakow, Poland ** AGH University of Science and Technology, Faculty of Computing Science, Electronics and Telecommunications, Department of Electronics, Al. Mickiewicza 30, 30-059 Krakow, Poland *** AGH University of Science and Technology, Faculty of Electrical Engineering, Automatics, Computer Science and Biomedical Engineering, student of Computer Science, Al. Mickiewicza 30, 30-059 Krakow, Poland **** The work was supported by the AGH-UST grant 11.11.120.612 (first author) and the AGHUST grant 11.11.230.017 (second author) A comparison of the SVM and Neural Networks in signal and image classification **** Introduction Most of the papers that dealt with comparing NN and SVM ended with a conclusion about the superiority of the relatively new technique of the SVM over the NN. The presented paper negates that thesis and depicts that in each presented example backpropagation neural networks, despite their unquestionable faults, is not worse than the currently fashionable SVM (Support Vector Machine) method and even quite often surpasses it. Neural networks managed quite well with an artificially generated problem representing a wide issue of segmentation of crucial objects from the background. The SVM method behaved unstably and reached an acceptable error level with difficulty. This additionally proves advantage of neural networks. It’s hard to think that a set of vectors was too small as 2000 elements were generated, one thousand for each of the two classes. As it is widely known an SVM classifier is unstable on a small-sized training set and SVM's optimal hyperplane may be biased when there are a lot fewer positive feedback samples than the negative ones. Two of the chosen examples had been soundly tested before and the test results had been published just for the neural networks. The earlier work based upon the older versions of the program STATISTICA by StatSoft so a repeated search for optimal networks was performed for the current comparisons using STATISTICA 10 software version. The results for the SVM method were also obtained using the above mentioned STATISTICA 10 software. Moreover, those results were verified in the application performed within the master thesis written by one of the co-authors of this paper. The research was held in the Laboratory of Biocybernetics and Biomedical Engineering. TOMASZ KRYJAK DAMIAN KROL MAREK GORGON 1 AGH University of Science and Technology, al. Mickiewicza 30, 30-059 Krakow tomasz.kryjak, ago@agh.edu.pl 2 damians.krol@gmail.com SEGMENTATION OF DISHES FOR THE PURPOSES OF CUSTOMER SERVICE PROCESS AUTOMATION IN A SELF-SERVICE CANTEEN Abstract. The article describes research on dishes segmentation for the purposes of customer service process automation in a self-service canteen. The project assumptions and a prototype test stand are presented. Also the developed empty workspace detection and tray position determination algorithms are discussed. Finally, the chosen dishes segmentation algorithm is described and justified. The article describes research on dishes segmentation for the purposes of customer service process automation in a self-service canteen. The project assumptions and a prototype test stand are presented. Also the developed empty workspace detection and tray position determination algorithms are discussed. Finally, the chosen dishes segmentation algorithm is described and justified. Piotr Pawlik, Zbigniew Bubliński* * AGH University of Science and Technology, Faculty of Electrical Engineering, Automatics, Computer Science and Electronics, Department of Automatics and Department of Automatics and Biomedical Engineering, Laboratory of Biocybernetics Image registration using Exponential Chirp Transform Abstract. The of aim article is to check the possibilities of the use of Exponential Chirp transform (ECT) instead of Mellin-Fourier transform (MFT) in areas related to matching patterns problem. In the paper a computational complexity of the algorithm based on ECT in relation to the MFT is estimated. Then the tests for shifted, scaled and rotated images are presented. JAROMIR PRZYBYŁO MIROSŁAW JABŁONSKI 1 AGH University of Science and Technology, Department of Automatics and Bioengineering Mickiewicza Ave., 30, 30-059 Kraków, Poland przybylo@agh.edu.pl , mjk@agh.edu.pl RAY-TRACING-BASED EVENT DETECTION AND 3D VISUALIZATION FOR AUTOMATED VIDEO SURVEILLANCE SYSTEM Abstract. Automated and intelligent video surveillance systems play important role in current home care and facilities security applications. Among many research problems is graphical visualization of semantic messages to the human operator that he can percept information in more natural way. The other essential research question is how to recognize 3D objects and their state on the monitored scene only from their views (2D images from the camera). In this paper we continue our previous work on data fusion in visualization of 3D scene semantic model and propose to recognize events and states of scene objects under surveillance in an automatic way using feedback provided by the renderer. We developed ray-tracing based visualization for surveillance system, that is capable of recognizing object’s state and at the same time present relevant information to the human operator. Automated and intelligent video surveillance systems play important role in current home care and facilities security applications. Among many research problems is graphical visualization of semantic messages to the human operator that he can percept information in more natural way. The other essential research question is how to recognize 3D objects and their state on the monitored scene only from their views (2D images from the camera). In this paper we continue our previous work on data fusion in visualization of 3D scene semantic model and propose to recognize events and states of scene objects under surveillance in an automatic way using feedback provided by the renderer. We developed ray-tracing based visualization for surveillance system, that is capable of recognizing object’s state and at the same time present relevant information to the human operator. Sławomir Mikrut* Agnieszka Moskal* * {Ph.D., Ph.D student} in Department of Geoinformation, Photogrammetry and Environmental Remote Sensing, AGH Kraków {smikrut, amoskal} @agh.edu.pl Selected Aspects of Processing Digital Images Obtained From UAVs Introduction The paper reviews current issues related to UAV flights. The relevant research has been conducted all over the world for years in various scientific research centres. The paper presents selected aspects of processing digital images made with the UAV technology. This paper provides a comparison between a digital image taken from an airborne height and the one made from an UAV level. Results of research show differences in images prepared with similar pixel sizes. The AGH University of Science and Technology in Kraków Department of Geoinformatics,Photogrammetry and Environmental Remote Sensing focuses mainly on geometry and radiometry of recorded images. The UAV technology seems to be highly future-oriented due to its low costs as compared to traditional aerial images taken from classical photogrammetry aircrafts.

Abstract The aim of this work was to develop a simple and fast yet robust algorithm for segmentation of objects in log-polar space. Assuming that some interesting parts of an image are selected by fixation of point of attention, pixels located close to fixation points are transformed to a log-polar space, then edge detection is performed. As a final step, detected edges are processed in a specific manner, which should lead to obtain boundaries of objects. After some smoothing, the detected boundaries are transformed back to a Cartesian space and are superimposed on an original image in order to show results of segmentation.

Abstract The aim of this work was to develop an algorithm for estimating the waiting time of the cars stopped before the intersection in a traffic flow measurement system (based on optical flow), which does not require the generation of the background and allows to calculation in real time. The proposed method performs analysis in HSV color space - a mask generated from S component is applied to H component. In this way a background - an asphalt and horizontal whitespace - is eliminated. The result of this operation is combined with data from optical flow to detect the vehicles which should be tracked.

Abstract The paper presents the implementation of modified background generation algorithm based on analysis of the frequency of occurrences of pixels. The proposed solution allows the use of the algorithm in real-time systems, which was quite difficult in the original method. The results show that the modified algorithm gives practically the same background as the original method but in significantly less time and can be applied in many tasks related to the detection and analysis of moving objects.

Abstract The paper presents the modification of background generation algorithm based on analysis of the frequency of occurrences of pixels. The proposed solution allows the generation of the background and its updating, the introduced parameter allows to adjust the algorithm according to the time rate of changes in the image. The results show that the modified method can be applied in many tasks related to the detection and analysis of moving objects.

Abstract The paper presents an attempt to create a system for visual surveillance for movement rehabilitation. The Delaunay triangulation was used as a key method for skeletonization of detected human silhouette. Also, the Hough transform was performed on skeletons of selected body parts in order to obtain slants of those body parts. Analysis of slant changes gives information of movement range, number of exercise repetitions and in general, to assessing correctness of performed exercises. The presented results show that - under some circumstances - it is possible to create visual system for such task.

The aim of the present study is to analyze ERCP images (endoscopic retrograde cholangiopancreatography). This means images of a contrasted pancreatic duct obtained through an X-ray method. More precisely, this involves creating an algorithm performing the first stage of the analysis/separation of the object of interest.