Por favor, use este identificador para citar o enlazar este ítem:
http://cimat.repositorioinstitucional.mx/jspui/handle/1008/1039| RGB-D Near Online Multi-Object Tracking (NOMT) | |
| ORLANDO GARCIA ALVAREZ | |
| Acceso Abierto | |
| Atribución-NoComercial | |
| MATEMÁTICAS INDUSTRIALES | |
| Tracking pedestrians and other moving objectives over RGB-D data is increasing in popularity, due to the access of cheaper and more precise depth devices. Keeping a consistent identity of these objectives over time is a combinatorial problem that, just like detection itself, is susceptible to occlusions, non-smooth movement, and even more so when the nature of a target so dynamic as a human being makes it difficult to maintain a consistent appearance measurement without using modern GPU-based techniques. In the literature, few tracking strategies have been designed to fully exploit 3D data and keep operation in real-time performance. In this thesis, we present a tracking pipeline that manages to quickly solve data-association on targets through a graphical model inspired by W. Choi’s NearOnline Multiple Tracking (NOMT) [CVPR15], with a redesigned energy optimization model for better usage of depth and color data and with the use of a simple histogram-based descriptor which is low cost for the processor and copes well with non-static targets. Along with this strategy, we present a new easy-to-use tracking software library that is compatible with ROS, providing code for faster RGB-D based tracker development. We also present an implementation and evaluation of our RGB-D NOMT with tests made over ground-truth data taken with Kinect sensors. | |
| 15-12-2018 | |
| Tesis de maestría | |
| OTRAS | |
| Versión aceptada | |
| acceptedVersion - Versión aceptada | |
| Aparece en las colecciones: | Tesis del CIMAT |
Cargar archivos:
| Fichero | Descripción | Tamaño | Formato | |
|---|---|---|---|---|
| TE 764.pdf | 3.08 MB | Adobe PDF | Visualizar/Abrir |