Institutional Repository of Vilnius University
Interaktyvus lazerio valdymas remiantis dirbtinio intelekto pozos atpažinimo duomenimis /
Abstract
dc:descriptionThis work explores the potential of a laser projector to project human movements using pose estimation tools. The theoretical part of the paper presents the working principles of laser projectors, introduces the International Laser Display Association (ILDA) Image Data Transfer Format and the applications of laser projection technology. Additionally, technical problems encountered by the projectors and the solutions to address these issues are presented. The subject of human pose recognition is also introduced and finally the theoretical part ends with a basis for real time systems. Before the experiment part, the hardware and software components of the project are introduced. The challenges of capturing video footage of laser projections are discussed and a solution involving optimized camera settings is introduced to mitigate any issues. The experiment evaluates two pose detection methods. The first is Google’s MediaPipe framework version 0.10.20 and its Pose Landmarker solution. The second method is the YOLOv11n-pose model, the latest version in the Ultralytics YOLO (You Only Look Once) series. Both models were tested in real time pose estimation tasks on different systems. The output data from the models was prepared and transmitted via a Helios Laser digital to analogue converter (DAC) to a laser projector using an ILDA connection. The conclusions are that projecting poses with a laser projector is possible even on older computer systems, but performance improves with newer hardware. The MediaPipe Pose Landmarker is more suitable for real-time applications on a central processing unit (CPU), providing faster frame rates in well-lit environments when the subject faces the camera, while YOLO performs better in complex conditions such as low light or the presence of obstacles. Future work recommendations include combining YOLO for human detection and MediaPipe for pose estimation within detected regions. This approach could enhance the ability to recognize multiple individuals in visual data while maintaining fast pose recognition. Additionally, using a video camera capable of recording at a higher frame rate with a progressive scanning mode could reduce the theoretical maximum DAC data transfer delay, improving overall system accuracy. Employing a separate camera for screen capture could also reduce CPU load, freeing up resources for pose estimation models and ensuring more stable and accurate results.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Tuchtenhagen, Lukas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD229556020&prefLang=en_US