| MAIG 2013 |
  |
|
 |
Dia amb activitat |
 |
Dia actual |
|
Segueix-nos a:
|
|
 |
MOBOTM: Mobile Robots Monitoring
|
01|01|1996
|
Data Inicial: 01|01|1996 - Data Final: 01|01|1999
Director d´investigació:
Jordi Vitrià
Membres:
Daniel Ponsa, Jordi Vitrià
Jordi Vitrià|Daniel Ponsa Client/Institució:
CVC - Centre de Visió per Computador
Duració: 3 years
Financiació: CeRTAP (Centre de Referència en Tecnologies Avançades de la Producció, Generalitat de Catalunya)
|
Design and development of a vision-based station for detection, tracking and precise localization of mobile robots in an industrial environment.
This project is devoted to the development of a monitoring station for visual detection and localization of autonomous mobile robots (or AGV´s) in an industrial environment. The monitoring station constitutes a low cost system for controlling systematic errors in odometry in the navigation system of the mobile robots as well as to annotate robot behaviors.
The system is composed of a video camera with a wide-angle lens, which provides detection and tracking facilities, and a video camera with a 25mm calibrated lens which estimates (precision: +/- 1cm) the robot position with respect to a map.
Visual monitoring is performed by an agent-based software system that implements real-time robot detection, tracking and behavior analysis in a standard PC station.
Motion detection is based on a statistical modelling of the environment. When the systems detects significant changes in local statistical descriptors, a tracking agent is assigned to that zone in order to follow the robot trajectory.
Robot motion is estimated assuming an affine model and using a Bayesian version of the Expectation Maximization (EM) algorithm to estimate model parameters. A deformable geometrical prior is used to impose a geometrical constraint on the optical flow distribution. This strategy results on robust estimation of robot movements even when the analysis is performed in noisy environments or in the presence of multiple targets. In this way the system can determine when the robot is visible for the calibrated system, and in that moment activates a measuring agent.
Robot position measurement is based on the detection of a set of visual landmarks designed on the robot surface. These landmarks, which are rotationally invariant, can be detected in real time using correlation operators.
This prototype is constructed using standard components, and shows that it is possible to design and construct complex visual systems using off-the-shelf hardware and software.
|
|
|
|
|
