Fryčák Tadeáš

Video processing

Flow velocity analysis

Pulpers mix the paper substance in their container. It is crucial to have a pulper that quickly gets the material from the surface down to the mixing blade. A lower blade speed causes no mixing on the surface at all, while a higher velocity leads to tedious circling of recently added material on the surface, which increases overall mixing time and energy consumption. To make the right choice of pulper, many other factors must be taken into account - container size and shape are just another two examples.

To set these properties correctly, we have developed a precise analytical tool for video comparison. We placed high-speed camera above the pulper to record flow of the liquid on the surface during mixing. Then our computer vision analytical tool comes into play, extracting speeds in each area on the surface from this video. We visualize this information using colors with a scale converted to meters per second (right video).

Finally, we simulate addition of new material to the surface (red circles coming towards the blade) and visualize them using streamlines (left video). All this information is subsequently used to select the best pulper correctly and precisely.

Image processing

Root Tracker

Root Tracker is a deterministic model for tracking roots without the need for extensive data annotations, as is necessary, for example, with convolutional neural networks.

For the purposes of the project, I have developed two independent computer vision algorithms that detect parts of roots from images regularly taken with a DSLR camera or a scanner at very high resolution. Subsequently, I devised an algorithm capable of automatically merging, tracking, and predicting individual roots, even when they overlap with the roots of other plants present in the same image.

Přírodovědecká fakulta Univerzity Palackého

Worm Skeleton

Deterministic methods for identification of objects contours often fail in complex microscopy images. It is where convolutional neural networks excel. They allow reliable finding of bounding boxes containing target objects and their classification even in images with uneven light and various artefacts. On the other hand, comparing of the bounding box positions between sequential images allows only very crude quantification of worm movement.

To solve these problems, Worm Skeleton combines both approaches. Bounding boxes of worms are predicted by neural networks and a sequence of deterministic methods is then applied within the individual boxes. This approach strongly reduces complexity of the problem and thus increases precision and saves computational resources.

At present, Worm Skeleton works with Caenorhabidits elegans, Haemonchus contortus and reasonably well with Teladorsagia circumcincta.

ALFA

Seven scanning flights were performed over 3 blocks of experimental barley field plots between April and June 2021. Resulting point-clouds were processed by the new fast algorithm ALFA for the processing of UAV LiDAR derived point-clouds to extract the information on crop height at many individual cereal field-plots at multiple time points.

The software converts point-cloud data into a digital image and extracts the traits of interest – the median crop height at individual field plots. Three different ways of crop-height data visualization are provided by the software to enable further analysis of the variability in growth parameters.

Web creation

Traion app

Web portal for internal communication and user management for the Traion network. The application is useful for welcoming new members, telling them what they can do in their first few days, or showing them who they can message. The administrator account then has permission to add events to the calendar for all other members, add materials for studying, or add verified suppliers. Everything was programmed using the latest technologies and arranged within an intuitive design.

3D

3D visualization

In the engineering and construction industry, visualizations play a crucial role in conveying design concepts, evaluating construction performance, and facilitating communication among stakeholders.

To achieve high-quality visualizations, we use top-of-the-line rendering software like Cinema 4D, which offers advanced capabilities such as realistic material and lighting simulation and camera animation. Our team of experts works closely with clients to understand their project requirements and develop custom visualizations that accurately depict the final product. Our visualizations enable clients to make informed decisions and showcase their designs to investors and the public.

Spindle app

The spindle apparatus is a cellular structure essential for the nuclear division of cells (mitosis or meiosis), ensuring the even distribution of chromosomes into daughter cells. However, errors can occur during chromosome division, which can result in serious developmental disorders (such as Down syndrome) or infertility.

The process of cell division is recorded using a so-called light-sheet microscope. This creates a time-lapse recording consisting of hundreds 3D cubes. Each cube then represents an individual oocyte at a specific time.

Based on these requirements and the data we received from the Institute of Animal Physiology and Genetics of the Czech Academy of Sciences, we have developed a application called Spindle App, which allows users to automatically segment the recordings in bulk. We decided to implement multiple segmentation models primarily to provide options for evaluation speed and precision preferences.