Resumen de Sistema de posicionamiento absoluto de un robot móvil utilizando cámaras externa
- español
En este trabajo se presenta nueva propuesta para el posicionamiento y guiado de robots móviles en espacios interiores, utilizando cámaras estáticas ubicadas en el entorno de movimiento y tratando de minimizar la infraestructura a bordo de los robots, La solución planteada está en la línea de lo que se conoce como "espacios inteligentes". donde la "inteligencia" del sistema se traslada al entorno, siendo éste el encargado de captar y procesar toda la información mecesaria àra pbtener la posición de los robots y generar las correspondientes órdenes de control, que les son enviadas vía radio, en función de las trayectorias que se deseen seguir.
Para la identificación de los robots se ha propuesto la utilización de cuatro balizas circulares dispuestas en forma de "T" y construidas a partir de diodos de infrarrojos. Estas balizas ubicadas a bordo de los robots solamente facilitan su localización, incluso en condiciones extremas de iluminación ambiental, sino que también simplifican los algoritmos de procesamiento de imágenes.
Las coordenadas 3D de la posición y orientación de los robots en el entorno se obtienen a partir de la información de las cámaras que observan la baliza, y es una de las aportaciones más importantes de este trabajo. La propuesta de posicionamiento se basa en minimizar un índice de comportamiento que tiene en cuenta la varianza de error en la determinación de las coordenadas de los centroides de la baliza en las diferentes cámaras. Para el caso de que los robots sean vistas por una única cámara, se utiliza el dato de la altura del plano en el que se encuentran la baliza sobre ellas para la obtención de las coordenadas 3D.
Para la segmentación de la baliza y obrención de su centoide, se ha utilizado operadores tipo gradiente que permiten una fácil y robusta detección de los bordes de la misma. A partir de las imágenes de bordes se presenta una umbralización dinámica basada en el método de Otsu. Teniendo presente que debido a la perspectiva, la baliza se proyectará como una elipse en las imágenes, para su localización se utiliza el método de la distancia algebraica que permite obtener las elipses candidatas que más se aproximan a cada uno de los objetos segmentados. A partir de las elipses candidatas se aplican criterios de excentricidad y relación de perímetros, para descartar falsos candidatos.
Se ha desarrollado un sistema de guiado utilizando Splines cúbicas para la generación de trayectorias. La posición de los robots se obtiene a partir de la información del sistema de visión, añadiendo una predicción y corrección por medio del Filtro de Kalman.
Este sistema novedoso de guiado permite que sea el entorno quien controle y dirija los movimientos y trayectorias de los robots, con la supervisión de su posición y estado en todo momento.
- English
This Thesis proposes a new method for the positioning and guidance of indoor mobile robots, employing static cameras which are located in the surrounding environment and minimizing the equipment on the robotic p!atforms. The research field of the proposed solution belongs to what is known as “intelligent environments”. The externa! systems to the robot acquire and process the information in order to obtain the robot position and give the commands for the robot guidance.
The main problems of the given solution are robot location and obtaining the 3D coordinates and orientation. The robots are identified by means of the detection of a structure of four circular beacons arranged in “T” shape. These beacons are constructed with infrared diodes. Besides detectingthe robot inpoor illumination conditions also al!owreducing the complexity of the algorithins of image processing. The position and orientation of the robots are extracted from the centroid of the beacons.
In the thesis a detailed study of the different error sources that have influence in the pixel coordinates obtained for the centroid of the beacons has been carried out. These errors have an important incidence obtaining the 3D coordinates of the robots. In this sense, different alternatives based on image processing have been proposed to minimize their effects.
One of the most important contributions of this thesis is the calculation of the 3D coordinates of the robots from the information obtained by the cameras capturing the beacons placed on board the robot. In this sense, when the robot is seen for severa! cameras, the proposal that has been carried out it is based on minimizing a criterion that considers the error variance of the centroids coordinates of the beacons captured by the different cameras. This proposal has demonstrated to give excel!ent results in empirical tests. In the case that the mobile robots are captured by a single camera, the a priori data corresponding to the height of the beacons platform over the robots is used in order to obtain the 3D coordinates. The robot orientation is computed considering the four beacons placed a T shape.
The difference in the luminance level between the beacons and background permits to carry out a segmentation which uses a processing algorithm based in the image gradient. Using a dynamic threshold (Otsu’s method) aliows for an easy and robust detection of edges Afterwards, keeping in mmd that the beacons were projected in the images as ellipses, an ellipsis locating method is proposed. From the ellipses candidates a criterion based on theblobs eccentricity, ratio of perimeters and geometric location among the four beacons is applied to discard false candidates. The practical tests have demonstrated a great reliability and robustness of the proposed method. After detecting the beacons in the image, a new alternative to improve the determination of their centroids has been developed. This proposal is based on considering as centroid coordinates the intersection of the main diagonais that connect the vertices of the minimum rectangle that circumscribe the beacon. This proposal provides high satisfactory results, and presents the advantage of lower computational cost than actual alternatives which have similar results from the point of view of our application.
Due to the great importance of camera calibration for a system of this type, a multi-camera calibration system has been developed considering two phases. First of ah the intrinsic parameters for each camera are obtained using an active pattern, also proposed in this thesis.
Second by, with the cameras located in their definitive location within the environment, the parameters of the rotation-transiation matrix between the different cameras are obtained. With this method, the errors for the intrinsic parameters when the calibration pattem is positioned away from the cameras are avoided. The practical results obtained with this alternative of proposed calibration have shown to be highly satisfactory.
In this thesis a system of mobile robots guidance has also been developed using cubic Splines for path generation. The position of the robot is obtained from the information of the multi camera system, adding a prediction and correction positioning values through a Kaiman filter. The prediction of the position of the robots permits to estimate the location of the beacons on board the robotic platform and consequently determine a region of interest to apply inside the algorithms responsible ofthe detection of the beacons centroids. This temporal and windowing filtering eliminates a large mount of noises and at the same time reduces the time of calculation of the algorithms.
Finally a practical guidance system of a mobile robot has been implemented, inside the building of the Escuela Politécnica Superior of Alcala University, making use of the proposed strategies in this thesis. The results of these tests have been highly satisfactory and have validated thus the developed proposais showing that it is possible to carry out the guidance of mobile robots using fixed cameras located in the surrounding environment, with all the advantages that it involves.
