Projects

This section presents a collection of projects and papers, central to my PhD research, unified by the theme of human pose estimation and analysis. The work spans various applications, from forensic tools for image and video processing in criminal investigations to modeling infant sensorimotor development and tracking athletes in sports. These efforts reflect a blend of technical innovation and practical application, emphasizing the importance of accurately analyzing human poses and movements to derive meaningful insights across diverse fields.

The project, a collaborative effort with VUT (Brno University of Technology), involves creating a suite of advanced forensic tools for the Czech Republic Police, funded by the Ministry of Interior. These tools focus on the automatic processing of images and videos, specifically honing in on human figures to enhance the efficiency and accuracy of criminal investigations. For more information about this innovative project, please feel free to reach out through any of the available contact methods.

The project led by Matej Hoffman, focuses on the behavior of infants to gain a deeper understanding of human development. My contribution to this project involves the precise estimation of 2D poses of babies in videos, a crucial aspect that helps in analyzing and interpreting infant movements and interactions. This work is part of a larger effort to model human behaviors, particularly in the early stages of life, providing valuable insights for various applications.

This project extends my long-standing interest in sports analysis, a journey that began with my bachelor thesis and evolved through my master's work. As a coach in the highest league of floorball, my focus in this project is on developing sophisticated methods for tracking floorball players in videos. The aim is to harness video analysis to gain insights into player movements and team dynamics, enhancing coaching strategies and game understanding. This work not only aligns with my academic pursuits but also integrates my practical experience and passion for floorball, striving to bring a new level of analytical depth to the sport.

The project led by Lukas Picek from University of West Bohemia focuses on the identification of wild animals (especially lynxes) in camera trap images. My contribution to this project involves the estimation of 2D poses of animals, which improves identification accuracy as each animal has a different texture on each side. The project is part of a larger effort to protect the environment using computer vision and AI.

A lightweight, local alternative to CVAT and LabelStudio. Originally developed to create the RePoGen dataset, we have since used it for multiple datasets. This easy-to-use Python tool features a simple GUI for annotating 2D human poses in images. Ideal for researchers and developers, PoseAnnotator simplifies the data labeling process for human pose estimation projects. Feel free to use it for your research and contribute to its development. Your feedback and contributions are welcome!

SKV Revie is a system for instant download of videos from the sports hall camera system. After pressing a button, the last 30 seconds of footage are saved, so you can download clips immediately after an interesting situation happens. The user interface is accessible from any device with a web browser, while the backend runs on a local mini PC. The system runs with latency below 0.5 seconds and can simultaneously handle (at least) 6 cameras in 4K resolution. We use Revie during both practices and matches in our floorball club. The system allows coaches to immediately analyze recent situations and provide feedback to players right after the end of a shift. The system is currently closed-source for internal club use. If you are interested in Revie, feel free to contact me — I am open to potential collaboration.

An automated system for downloading videos from SKV's sports hall. The system interface is available on the public website, accessible only for SKV staff. It allows downloading videos from all our cameras with 30-day history. The backend runs on a local server and is connection of Python backend, Microsoft Power Automate and Microsoft OneDrive. The Python backend logs every activity and sends failure notifications to the developer (me) to allow almost real-time support. So far, the system downloded more than 400 videos.

I was part of Porsche's team developing the world's most powerful EV charger, a breakthrough in charging speeds and efficiency. Our team was responsible for developing the software across all system components. For more details on this innovative technology, see [this article on 800-volt charging](https://www.smart-emotion.de/article/405-800-volts-explained-why-porsche-redefined-charging-times-and-efficiency/).