REUNIÓN DEL GRUPO DE VISIÓN


Jueves 5 de Septiembre


11:30 a 12:20 Conferencias invitadas de jóvenes investigadores españoles
Arnau Ramisa: Using depth and appearance features for informed robot grasping of highly wrinkled clothes
Ana C. Murillo: Place Recognition and Scene Understanding with Panoramic Images

12:20 a 13:15 Conferencia invitada
Margarita Chli (ETH Zurich): Visual SLAM for Micro Aerial Vehicles

13:15 a 13:45 Charla técnica INFAIMON
David GuillametSistema de bin-picking 3D basado en SLAM visual

13:45 a 14:15 Reunión del grupo temático de visión

14:15 a 15:00 Comida de trabajo durante la reunión del grupo temático (se comerá en la misma sala)

15:00 a 15:30 Cierre de la reunión del grupo temático de visión


ARNAU RAMISA

Title

Using depth and appearance features for informed robot grasping of highly wrinkled clothes


Abstract

Detecting grasping points is a key problem in cloth manipulation. Most current approaches follow a multiple re-grasp strategy for this purpose, in which clothes are sequentially grasped from different points until one of them yields to a desired configuration. In this paper, by contrast, we circumvent the need for multiple re-graspings by building a robust detector that identifies the grasping points, generally in one single step, even when clothes are highly wrinkled. In order to handle the large variability a deformed cloth may have, we build a Bag of Features based detector that combines appearance and 3D geometry features. An image is scanned using a sliding window with a linear classifier, and the candidate windows are refined using a non-linear SVM and a "grasp goodness" criterion to select the best grasping point. We demonstrate our approach detecting collars in deformed polo shirts, using a Kinect camera. Experimental results show a good performance of the proposed method not only in identifying the same trained textile object part under severe deformations and occlusions, but also the corresponding part in other clothes, exhibiting a degree of generalization.

Bio

Arnau Ramisa is a postdoc at the Spanish Scientific Research Council (CSIC). He received the MS degree in 2006, and the PhD degree (with highest honors) in 2009 from the Autonomous University of Barcelona.
In 2010 he was a postdoctoral researcher with the LEAR group in INRIA Grenoble. Since 2011, he has been with the Institut de Robotica i Informatica Industrial (CSIC/UPC), Barcelona. His research interests include object classification, detection and segmentation applied to robot vision.


ANA C MURILLO

Title

Place Recognition and Scene Understanding with Panoramic Images

Abstract

The need for mapping and modeling larger environments requires models at different levels of abstraction and additional abilities to deal with large amounts of data efficiently. Omnidirectional vision systems are of particular interest because they allow us to have more compact and efficient representation of the environment. I will present two of our projects results using omnidirectional imagery: place recognition on large outdoor environments and semantic labeling of places indoors using wearable devices.

Bio

Ana C. Murillo obtained her PhD in Computer Science at the University of Zaragoza, Spain, in 2008 as a member of Robotics, Perception and Real Time group. Since then, she is a researcher and assistant professor of the same institution. Her current research interests are in the area of computer vision, in particular, place recognition, semantic mapping and scene understanding, with special interest in omni-directional vision systems and applications to robotics and other mobile and wearable devices.


MARGARITA CHLI

Title

Visual SLAM for Micro Aerial Vehicles

Abstract

Simultaneous Localisation And Mapping (SLAM) forms the backbone of spatial awareness for autonomous systems. With the enormous descriptability of images of a robot's surroundings, visual SLAM is at the heart of bridging the gap between the Computer Vision and the Robotics communities. However, the wealth of visual data together with the requirements for true robustness and the hard real-time constraints in robotics systems, pose great challenges for truly general navigation systems.
In this talk, I will present how we address these challenges and our latest results towards fully autonomous vision-based flights using Micro Aerial Vehicles. The onboard sensing and limited computational capabilities, resemble those of a smart phone, however, the great agility of MAVs poses unique challenges e.g. in visual tracking. It is only after we have successfully addressed these problems that we can make effective use of such systems in tasks like search-and-rescue.

Bio

Margarita Chli is currently Deputy Director of the Autonomous Systems Lab (ASL) of ETH Zurich, Switzerland. She received the B.A. and M.Eng degrees in Information and Computing Engineering from the University of Cambridge, UK. In 2006, she became a member of the Robot Vision group of Imperial College London, London, UK, where under the supervision of Prof. Andrew J. Davison, she worked on applying Information Theory for efficient Simultaneous Localization And Mapping (SLAM) using vision, to receive the PhD degree in 2009. Building on her thesis work, she continued at the Robot Vision lab as a Research Associate until October 2010, when she joined the ASL of ETH Zurich as a Senior Researcher and Lecturer. From January 2013, she is appointed Deputy Director of the ASL. Her research focuses on robust vision-based SLAM on limited computation platforms like micro aerial vehicles, leading the group’s team for the European projects sFly, myCopter and SHERPA.


DAVID GUILLAMET

Departamento I+D de INFAIMON

Título

Sistema de bin-picking 3D basado en SLAM visual.

Contenido

Se define como bin-picking a la tarea robótica de coger (pick) todas las piezas de un contenedor (bin) en el que están colocadas de forma aleatoria, para dejarlas en una localización conocida de modo que puedan ser utilizadas en un proceso productivo posterior. Actualmente esta tarea se realiza de forma manual, por lo que existe una gran demanda en la industria para este tipo de soluciones. Para resolver el problema de bin-picking hay que llevar a cabo la detección de objetos en 3D con 6 grados de libertad (x, y, z,roll, pitch, yaw) y tratar con problemas como oclusiones, solapamientos, problemas de foco, escala, variación de formas, iluminación no controlada. Todo ello con robustez industrial.

Información adicional