Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/7F7D3A66-EEC2-445E-9DCA-292387661722 | ||
| Year | 2024 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Dimitrios Ntounetas, "Design and implementation of an embedded system for stereo vision from a single camera", Diploma Thesis, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2024 https://doi.org/10.26233/heallink.tuc.98700 | ||
| Appears in Collections |
3D Vision has been a topic of investigation for numerous decades, with researchers striving to overcome the challenge of losing the third dimension when using images. Over the years, various techniques have been devised to capture depth from images, including stereo vision, fringe projection, laser scanning, or a combination of these methods. While these techniques have demonstrated impressive results, they may not be suitable for certain applications due to factors such as high cost. Consequently, there is a pressing need to develop a 3D vision system that is affordable, fast, and capable of producing high-resolution output. This thesis implements an innovative approach to estimate depth, initially studied by G. Rematska in her Master Thesis. Depth is calculated by using a single camera in conjunction with two spectrally distinct light sources. The light sources consist of two sets of LED arrays, and depth information can be extracted by analyzing the different reflections captured by the camera from these two light sources. The key aspect of depth estimation lies in the blue to red ratio, which can be correlated to depth through appropriate calibration and processing. In this thesis, the goal was to evolve a system which was verified under laboratory conditions (controlled lighting, no camera vibrations, completely fixed distance of camera to target surface) into one suitable for field use, in which the conditions (lighting, camera vibrations, distance to target surface) are widely varied. The methodology was verified using MATLAB and subsequently implemented as a real time embedded system. The resulting system was thoroughly tested in the field and operates in real time and is fully optimized to minimize hardware resource usage while maintaining a high frequency of operation for processing high-resolution images.
Copyright © DIAS 2013
