Seguimiento monocular 3D para rehabilitación
In this thesis, there are shown the results of the development of a three dimensional monocular tracking system that detects objects and can work even with low cost webcams, the system has application for in house and self directed rehabilitation. The developed system is able to detect and track col...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2008 |
| País: | México |
| Institución: | Instituto Nacional de Astrofísica, Óptica y Electrónica |
| Repositorio: | Repositorio Institucional del INAOE |
| Idioma: | español |
| OAI Identifier: | oai:inaoe.repositorioinstitucional.mx:1009/534 |
| Acceso en línea: | http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/534 |
| Access Level: | acceso abierto |
| Palabra clave: | info:eu-repo/classification/Visión por computador/Computer vision info:eu-repo/classification/Análisis de movimiento de la imagen/Image motion analysis info:eu-repo/classification/Seguimiento del objetivo/Target tracking info:eu-repo/classification/cti/7 info:eu-repo/classification/cti/33 info:eu-repo/classification/cti/3314 info:eu-repo/classification/cti/331110 |
| Sumario: | In this thesis, there are shown the results of the development of a three dimensional monocular tracking system that detects objects and can work even with low cost webcams, the system has application for in house and self directed rehabilitation. The developed system is able to detect and track color objects. In order to do it, the system requires a sample from which is obtained a color histogram. Then the color histogram is used to compute a backprojection over every image captured from the camera. The backprojection process gives a probability maps for every frame. Those maps indicate the presence or absence probability of the object. Then, the probability maps are fed to a Bayesian Occupancy Filter (BOF). The backprojection is inserted into the BOF grid, which in turn records the probability of each cell being “occupied”. Afterwards a clustering algorithm is applied over the grid to find clusters of highly occupied cells. At the end, the highly occupied clusters are marked as objects on the scene and are tracked in 2D. In this work we focus on tracking one object, so only the object with the highest certainty is tracked and the rest (if any) are dismissed. The depth detection is archieved via the knowledge of the real size of the tracked object and the fact that relative depths can be easily measured by using only one camera, also the discretization is important for the depth detection and some factors obtained via the BOF. Experiments were conducted to measure the precision of the tracking. The system was compared with a stereo system and similar performance was measured. The system has the advantage of less implementation complexity since it does not require camera calibration nor special mounting hardware as the stereo system does. It can be used with typical webcams. The system was integrated into a rehabilitation system named Gesture Therapy with promising results so pilot clinical tests were started. |
|---|