Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/9B66AB46-02E0-4447-9A03-D56AC8CF5150 | ||
| Year | 2022 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Veniamin Malefioudakis, "Real time object reconstruction in embedded systems", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2022 https://doi.org/10.26233/heallink.tuc.92674 | ||
| Appears in Collections |
The three-dimensional reconstruction of objects is evolving more and more in thethe field of machine vision. Through appropriate algorithms, it is now possible toproduce synthetic 3D scenes in real time. In this direction, the synthesis of realistic 3D models remains demanding and time-consuming, significantly increasing thecomputational cost. In this thesis, the main purpose is the development and implementation of a integrated application for 3D image reconstruction (disparity map) in an embedded system, using optical sensors at very low cost. The application is designed and implemented to have a graphical interface that allows the user to easily perform all the procedures required for the 3D reconstruction of the scene. In addition, it offers the possibility of selecting one of the two stereoscopic algorithms incorporated to create the depth map of the scene and to adjust their parameters in real time. The application has been optimally developed in terms of the efficient use of computer system power. The user can select either the computing processing unit (CPU) or the graphics processing unit (GPU) for the calculation of the 3D graphics in real time. The calculations and comparisons performed in this thesis are focused on the use of a compatible desktop PC and the use of the embedded system Nvidia Jetson NX Xavier (Jetson). Comparisons with input images of 640 or 1080 pixels have been made, between the two different algorithms, between the CPU and GPU and between the two computing systems (Desktop PC vs. Jetson). From these comparisons, it is shown that using Jetson for real-time disparity map computation is very close to the Desktop PC times (8 ms difference for GPU calculations and using 640 pixels and 100 ms for CPU calculations using 1080 pixels). From these results, it can be seen that the use of embedded systems with low optical sensors are now able to support portable and manned or unmanned systems with a wide range of applications such as airborne 3D space reconstruction.
Copyright © DIAS 2013
