Ing. Jakub Novák

Ing. Jakub Novák

Mgr. Jan Spěvák, Ph.D.

Department of Applied Mathematics

The bachelor thesis aims at the acquisition and subsequent detection and classification of surface defects in chocolate moulds used for the production of chocolate candies, etc. A sensing system using a line camera, dark-field illumination and bright-field illumination is proposed for image acquisition. The detection algorithm is based on edge detection. The defects are classified into two groups according to the test by which they are detected. The detected edge defects (first group) of the cavities are correctly detected in 78 % of cases. Surface defects (second group) were correctly detected in 76 % of cases.

Thesis on DSpace

Ing. Jakub Novák

Ing. Magda Friedjungová, Ph.D.

Department of Applied Mathematics

This thesis focuses on the automated detection and classification of defects on glass rods. Previous solutions of similar problems are analysed and a new approach is proposed. Four camera systems using advanced optical components and lighting are designed. Also, analysed defect detection and classification methods are modified and extended. One of them using a model based on the modern Faster R-CNN architecture is able to detect 83 % of defects and to correctly classify 77 % of them.

Thesis on DSpace

Ing. Jakub Novák

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Department of Applied Mathematics

The thesis is about extraction of text information from photographs of identity documents. Proposed method uses convolutional neural network U-Net trained to create segmentation mask to find ID card in natural scene. Text is detected within known areas and is read using Tesseract library. The character recognition precision is 98,4 % and recall is 92,6 % on 48 tested photos of documents.

Thesis on DSpace

Ing. Jakub Novák

Mgr. Eva Pernecká, Ph.D.

Department of Applied Mathematics

The thesis focuses on the analysis and evaluation of the technique of strengthening exercises using the methods of computer vision and machine learning. From the input videos of the exercisers, human skeletons were obtained and parameters describing the course of the exercise were calculated from them. Using the Dynamic Time Warping method, the tested exercises were compared with the ideal course and each parameter was evaluated separately. Different models for skeleton extraction were compared, as well as different positions from which the exerciser was recorded. For the selected orientations from which the exercises were filmed, the quality evaluation of the exercises reached average accuracy of 90 % and average f-score of 83 %.

Thesis on DSpace

Ing. Jakub Novák

prof. RNDr. Pavel Surynek, Ph.D.

Department of Applied Mathematics

This thesis is concerned with the mapping of limitations on creating point clouds and measuring depth using commonly available depth-sensing cameras. The influence of an incidence angle, measurement distance, object shape, size and surface characteristics were investigated and measured on three devices, utilising Time-of-Flight, passive stereo vision, and active stereo vision. The compared devices are Basler ToF, OAK-D-Lite, and Intel RealSense D415. The tests established the strengths and weaknesses of each device. The Basler ToF has the best structural resolution but is easily influenced by the surface characteristics of the measured objects. The depth accuracy and structural resolution of the Intel RealSense D415 are comparable to the Basler ToF but are not as easily influenced by a surface finish. The OAK-D-Lite lacks the structural resolution required for an accurate reconstruction of 3D objects. However, it seems to be a solid choice for an approximate scene depth estimation in visually complex environments. The results of the carried out experiments are usable as a basis for feasibility studies on computer vision projects utilising depth-sensing cameras.

Thesis on DSpace

Ing. Jakub Novák

Ing. Miroslav Čepek, Ph.D.

Department of Applied Mathematics

The paper compares algorithms designed for image segmentation. Three algorithms belonging to the category of superpixels (felzenszwalb, SLIC and quickshift), two representatives of active contour models (snakes and level sets), random walker, region adjacency graphs and Otsu thresholding are compared. For this purpose, defects appearing in the industry are described. A more general categorization splitting defects into groups is made. Subsequently, two defects are selected from each category. A suitable combination of algorithm parameters is found on the first defect. Different preprocessing and color representation of the image are accounted for when the search is carried out. With the found parameters, a segmentation of the second defect is performed. Thus, the generalization capability and the suitability of the algorithm for the defects of the given category are tested. The success of the segmentation is measured by the IOU metric. The more successful algorithms achieved an average IOU measured over all images of a single defect of 90 %. In generalization test an average IOU of 53 % was achieved in some cases.

Thesis on DSpace

Ing. Jakub Novák

Ing. Magda Friedjungová, Ph.D.

Department of Applied Mathematics

The thesis deals with the optimization of defect detection in NDT (non-destructive testing) methods. The problem is solved by designing a new detection algorithm, based on the U-NET model, which solves the shortcomings of the earlier model. Also, it is discussed, how to algorithmically expand data for training. The result of this work is an optimization of the process, which significantly increased the accuracy of the entire detection system. These optimizations are universally applicable to the problem of detecting small defects and can thus improve system performance even outside the NDT domain.

Thesis on DSpace

Ing. Jakub Novák

doc. RNDr. Pavel Surynek, Ph.D.

Department of Applied Mathematics

The subject of this thesis is automated detection of defect indication on parts inspected by methods of non-destructive testing, specifically magnetic particle testing. Previous solutions to this problem are analyzed and a new approach is proposed. First, a camera system is put together and adapted to the testing method. Then an indication detection algorithm, based on modern R-CNN methods, is designed and implemented. For the implementation Detectron2 system from Facebook is used. With a training dataset, which consists of 27 images and a test dataset consisting of 10 images, the model detects 83 % of all true indications, while making an error in every fifth image.

Thesis on DSpace

Ing. Jakub Novák

Mgr. Jan Spěvák, Ph.D.

Department of Applied Mathematics

The thesis deals with measuring the influence of archery technique on the arrow flight. It makes use of a high-speed camera to analyze the sportsman's technique. The archer's movements directly influence the trajectory of the arrow, which is presented by its position in the target. A recording of the hands at the moment of string release is made using a high-speed camera. After that, methods of image processing are used to acquire the positions of fingers from these captured frames. Together with every analyzed shot a position of the arrow is noted using a Cartesian coordinate system with the origin at the center of the target. It is confirmed, that short distance between the index and middle finger is the cause of arrows falling very far from the center, that greater horizontal distance of the hand from the anchor position leads to hits in the left side of the target and finally that a lower rotation of the hand leads to hits below the center of the target and vice versa.

Thesis on DSpace

Ing. Jakub Novák

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Department of Software Engineering

The focus of this thesis is on the size measurement of jewelry stones based on computer vision and image processing. The quality of jewelry stones is inspected by the size of its circumscribed and inscribed circle, the size of the circumscribed circle of the diamond's table, level of polish and center offset. Three acquisition systems and four algorithms are presented. The acquisition system is composed of a monochromatic camera, a telecentric lens, a coaxial light and a specific bar diffuser and ring diffuser. Algorithms are implemented using the Canny-Otsu detector, Otsu thresholding and watershed segmentation. The accuracy of the measurements is 0,0098 mm, size deviation related to the position in the image oscillates within a range of 0,0028 mm and 0,0040 mm.

Thesis on DSpace

Ing. Jakub Novák

doc. Ing. Štěpán Starosta, Ph.D.

Department of Applied Mathematics

The bachelor thesis focuses on the automatic evaluation of building kit parts placement using a mobile device such as a tablet. The building kit is a part of iSophi SMART project. First, measurement methodology is designed in the way the user is able to view the tablet's display, reach for building kit parts and still see the whole scene. Then an algorithm, which extracts information from the image about the building kit parts, is implemented by using traditional image processing methods such as mean-shift filtration, color thresholding, and marker-controlled watershed segmentation. Lastly, the algorithm is implemented in the form of a server application, which takes a request to validate an input image. Information from the image is compared to the reference solution of the task, and the result is returned as a response. The proposed application was tested on 116 images under two light conditions. The success rate under the daylight and the room light is 84,48 % and 88,79 % respectively.

Thesis on DSpace

Ing. Jakub Novák

Dr.-Ing. Martin Novotný

Department of Digital Design

The aim of the work is to create a hardware system to focus the view of industrial camera in the sense of getting higher resolution in area of interest. A solution to use two cameras was chosen. The first one is an overview camera and the second one is a detail camera, which gaze is aimed by a mirror tilted in two axis. The created solution enables to aim to a wanted object from an overview camera and retrieve its detail in higher resolution by the detail camera. The main result of the work is the created hardware system and its firmware to control tilting of the mirror.

Thesis on DSpace

Ing. Jakub Novák

doc. Ing. Štěpán Starosta, Ph.D.

Department of Applied Mathematics

The diploma thesis deals with precise 3D measuring techniques in industry use. The thesis is focused on the laser triangulation technique. The influence of scanned object material, laser triangulation geometry, the angle between camera and laser and laser color on the precision of measurement were analyzed. A universal acquisition system with swing arms was introduced for a laser profiler. Precision was analyzed using four experiments using the RANSAC algorithm. In some cases, correction of rotated profiles and RMSE metric for evaluation was applied. In material influence analysis, the best results were obtained in measurement of matte copper and wood. The maximal error in measured cube dimensions was 0.4 mm. Best precision when capturing a wooden ball was achieved with the laser pointing from above and the camera from the side, the angle between the camera and laser of 30 * and red laser color. The radius error was 1 mm, and RMSE metric 0.3458. The best laser profiler 3D measuring precision was achieved when scanning matte wooden objects using geometry 1 (laser pointing from above, camera from the side) with an angle of 30 * and red laser.

Thesis on DSpace

Ing. Jakub Novák

doc. Ing. Štěpán Starosta, Ph.D.

Department of Applied Mathematics

This thesis aims to automate advertisement detection and corresponding interactions done by a user from a screen recording of the Instagram iOS application. The solution consists of detecting a location, a new post, an advertisement, a type of media, and the actual interactions. Computer vision techniques, such as template matching, Hough circles, SSIM index, or segmentation by thresholding, are used to resolve each subproblem. Algorithms are implemented in the form of an application, which receives video, system skin, and type of the device as input and outputs a structured text file. The output file consists of transactions, which are composed of the advertisement, additional information such as the location of detection, length of interactions with the advertisement or its type, and interactions. On the test dataset, the proposed application achieved to detect the advertisement in 92,16 % of cases and corresponding interactions with a success rate of 93,53 %. Results on the dataset from user testing show 78,57 % correctly detected advertisements and interactions with a success rate of 81,25 %, however with a high rate of false positives.

Thesis on DSpace

Ing. Jakub Novák

doc. Ing. Tomáš Vitvar, Ph.D.

Department of Software Engineering

The goal is to design a system that can handle data from multiple vision systems. The solution is built on a client-server relationship. A total of 4 separate applications were created. The C++ language was used for the implementation. Functionality tests were successful. In terms of scalability, the solution can handle 30 data sources and process 100 MB of input data from 10 sources in an average of 4 seconds.

Thesis on DSpace

Ing. Jakub Novák

doc. Ing. Ivan Šimeček, Ph.D.

Department of Software Engineering

The diploma thesis concerns with the design and the implementation of Android mobile application as a medical decision support system. The analysis of contemporary medical applications and analytic tools was carried out, as well as the analysis of Alma Diagnostics web application. The web application is the basis for the mobile application. It also serves as the source of relevant data by means of architectonic style REST. The thesis further analyzes medical data standards and proposes their utilization.

Thesis on DSpace

Ing. Jakub Novák

Ing. Jitka Hrabáková, Ph.D.

Department of Applied Mathematics

The thesis deals with the measurement of heart rate using a conventional camera. The measurement process is divided into three main parts: face detection, skin segmentation and heart rate calculation. Face detection uses a single shot detector neural networks. Skin segmentation can be done either by traditional thresholding methods or by using a neural network to create a face mesh. Heart rate calculation is based on estimating the blood volume pulse from subtle changes in skin tone, followed by Fourier transform to determine the number of heart beats per minute. A camera system is designed and a configurable application is developed. The resulting accuracy in measuring resting heart rate (50-80 BPM) is comparable to reference devices with a measurement error of 5 %. At higher heart rates (80-200 BPM) the measurement error is 10 %.

Thesis on DSpace

Ing. Jakub Novák

Department of Applied Mathematics

This thesis focuses on automated detection of protein particles on microscope images. Distribution of the detected particles is analyzed. The thesis contains an analysis of existing solutions to similar problems and description of selected image processing methods. These methods have been implemented in Java and used in the design of a particle detection algorithm. Several method combinations have been tested on real data and compared to manually annotated data. Particle positions detected by the best algorithms have been processed by selected spatial analysis techniques.

Thesis on DSpace

Ing. Jakub Novák

doc. RNDr. Ing. Marcel Jiřina, Ph.D.

Department of Software Engineering

The diploma thesis concerns with the renewal of the DITHYRON program designed for thyroid disorder diagnostics during 1990s for MS-DOS operating system into a module form of contemporary Alma Diagnostics web application. The thesis contains detailed analysis of historical implementation shortcomings, which caused the program to be unsuitable for present usage, expansion and improvements. The thesis also analyzes present applications that are used to help with the disease diagnostics and problems that limit their expansion. Based on these findings a module for help with thyroid disorders diagnostics was designed and implemented.

Thesis on DSpace