Ing. Justýna Frommová

Ing. Justýna Frommová

Ing. Matěj Mráček

Department of Applied Mathematics

This bachelor's thesis focuses on the detection and classification of defects in paper machine fabrics, which are in many ways similar to technical textiles. The main goal was to design and assemble an imaging system capable of collecting image data, collect data, and develop algorithms for automatic defect detection and classification. The thesis describes the design of the optical setup, selection of appropriate illumination, and camera system. It also presents image processing methods including filtering, transformations, and segmentation techniques used in the detection process. The outcome is a set of algorithms that can identify and distinguish specific types of defects with a certain level of accuracy. The final part of the thesis discusses the advantages and limitations of the proposed solution and outlines possibilities for future development.

Thesis on DSpace

Ing. Justýna Frommová

RNDr. Luděk Kleprlík, Ph.D.

Department of Applied Mathematics

This thesis focuses on the realistic and visually appealing fusion of vehicle images with selected backgrounds using 2D image processing techniques. To enhance the credibility of the image synthesis, reflections and shadows are generated for the vehicle. Particular attention is paid to parameterization and control of output properties such as shadow intensity, light source position, and appropriate reflection deformation. As part of the implementation, an algorithm was developed based on the Pixel Height method for modeling shadow geometry, enhanced with soft shadows, vehicle reflection using column rotation with smoothing via a moving average, and seamless foreground-background merging through alpha blending and edge blurring. Results are presented in both virtual environments and a real car showroom. The output of the algorithm demonstrates that the proposed solution is capable, except for few limitations, of producing aesthetically and technically acceptable results suitable for marketing purposes.

Thesis on DSpace

Ing. Justýna Frommová

Ing. Jitka Hrabáková, Ph.D.

Department of Applied Mathematics

The thesis designs and implements an algorithm capable of automatically creating video highlights from surf competitions. It also provides a dataset of 47 hours of annotated surf videos. A classification model utilizing MobileNet and LSTM binary classifies video frames as containing or not containing a surfer riding a wave. Signal smoothing methods are applied to its predictions and moments containing a riding surfer are selected. From these, output in the form of an edited highlights video is created. The algorithm achieves an adjusted accuracy of 97 % and correctly identifies 88 % of waves surfed. It completely misses only 1.5 % of surfed waves.

Thesis on DSpace

Ing. Justýna Frommová

Ing. Jitka Hrabáková, Ph.D.

Department of Applied Mathematics

This thesis deals with non-contact measurement of pin heights in connectors. The Shape from Focus method is chosen for the measurement, which determines the depth of a surface from a series of images captured at different focus planes. The thesis describes the design of the imaging system, alignment of the image sequence, depth map calculation, pin segmentation, and determination of their heights. The solution also includes verification of the applicability of an electrically tunable liquid lens for changing the focal plane. Four focus metrics are implemented and compared. The best results are achieved using the Sum of Modified Laplacian metric. The results confirm the applicability of the proposed method for measuring pin heights.

Thesis on DSpace

Ing. Justýna Frommová

RNDr. Luděk Kleprlík, Ph.D.

Department of Applied Mathematics

This bachelor thesis deals with assisted measurement of the lateral distance of fixed roadside obstacles from monocular road network videos. The aim is to design and implement a tool applicable in the context of the CEBASS me- thodology, combining road reference line detection with subsequent geometric scene evaluation. The thesis compares three approaches to reference line de- tection: a classical baseline based on the Canny edge detector and the Hough transform, the UFLD-v2 model, and the transformer-based LSTR model. The distance measurement pipeline uses camera parameter estimation, a road plane model, manually selected obstacle points, and metric visual odometry from a short sequence of frames. On the validation data, LSTR proved to be the most suitable reference line detector. The final measurement pipeline was eva- luated on a test set of 18 cases, all of which were successfully processed; the mean absolute error was 0.79 m and the median absolute error was 0.47 m. The results indicate that the proposed solution can serve as an interpretable assisted tool for converting image data into metric distances, while remaining sensitive to the quality of reference data, lane marking detection, and the sta- bility of the geometric reconstruction.

Thesis on DSpace