COMPARAÇÃO ENTRE CONTROLES LOOK-AND-MOVE E SERVO-VISUAL UTILIZANDO TRANSFORMADAS SIFT EM MANIPULADORES DO TIPO EYE-IN-HAND / COMPARISON BETWEEN LOOK-AND-MOVE AND VISUAL SERVO CONTROL USING SIFT TRANSFORMS IN EYE-IN-HAND MANIPULATOR SYSTEMS

AUTOR(ES)
DATA DE PUBLICAÇÃO

2009

RESUMO

Computer vision can be used to calibrate and self-localize robots. There are many applications in self-localization and control applied to industrial manipulators and mobile robots. In particular, visual control can be useful in submarine interventions, where a robotic manipulator is mounted on a Remote Operated Vehicle (ROV) to execute tasks at high depths, such as handling manifold valves. This work has the objective to develop and implement visual control techniques to self-localize and position robotic manipulators. It is assumed that a monocular camera is attached to the robot end-effector (eye-in-hand configuration). Two visual control techniques are studied: look-and-move and visual servo control. Their main difference is related to the adopted feedback sensors. The first technique uses position sensors with the aid of a single image captured at the beginning of the robot movement. The second technique relies on several images captured in real time during the robot movement. The main contribution of this work is the use of the SIFT transform, robust to rotation, translation, changes in scale and illumination, to obtain and correlate key-points between reference images and images captured in real time. The methodology is experimentally validated using a manipulator based on the mechanical structure of an x-y-0 coordinate table. An electronic system was developed to control the robot through a software in a computer, where were implemented all the techniques proposed. Preliminary tests are performed on simple circular-shaped objects, without the need for SIFT transforms. Next, tests are performed with a photo of an actual manifold panel typically used in submarine interventions, using SIFT transform to find the localization of the manipulator with respect to the panel. The results show that the performance of the visual servo control depends on the image processing time, unlike the look-and-move. However, the visual-servo control presents smaller positioning errors. The SIFT method is appropriate for both controls, since image resolution be high enough to avoid false matching.

ASSUNTO(S)

computer vision visao computacional

ACESSO AO ARTIGO

Documentos Relacionados