SPEAKER:
Patrick McGuire
Centro de Astrobiología (INTA/CSIC)
Instituto Nacional Técnica Aeroespacial
Torrejón de Ardoz, Madrid, Spain

DATE & LOCATION:
20 April 2005, 12pm Rm. 203, EPSc
McDonnell Center for the Space Sciences, Depts. of Physics and Earth & Planetary Sciences
Washington University in St. Louis

ABSTRACT:
We honor the 400th anniversary of the publication of "Don Quixote de la Mancha" (1605) by Cervantes by using the title character and his exploits to illustrate the aspirations, journeys and improving-perception capabilities of our Cyborg Astrobiologist computer-vision system.

The Cyborg Astrobiologist's hardware consists of a raster-capable camcorder connected by a Firewire cable to wearable computer, together with a human operator.

The Cyborg Astrobiologist's software consists of an image-segmentation algorithm and a technique to search for uncommon regions in an image. We are using a graphical-programming language from Bielefeld, Germany, called "NEO" for developing the computer vision and autonomy software on the wearable computer.

With Don AstrobioQuixote (a.k.a. the Cyborg Astrobiologist) and his "armor" of a wearable computer and his "lance" of a camcorder, we present computer-vision exploration results from several missions at two different field sites in Spain, one with white gypsum cliffs and the other with red sandstone outcrops. The image-segmentation and uncommon-mapping technique that we developed for the Cyborg Astrobiologist's wearable computer performed admirably in the field, yet there is much room for improvement. Through a collaboration with researchers in Girona (Catalonia, Spain), we hope to improve the image segmentation in the coming months to handle color and texture simultaneously. Later this year, we also hope to use the image segmentation and the uncommon-mapping technique to autonomously guide a robotic borehole-inspection system as part of the NASA/Ames--Spain "Mars Analog" underground-drilling collaboration (M.A.R.T.E.), near the Rio Tinto in SW Andalucia, Spain.

These image-segmentation and uncommon-mapping techniques could be suitable for utilization after the Mars Reconnaisance Orbiter begins its studies next year, for the CRISM hyperspectral imager. Together with classification techniques, like neural networks and support vector machines, the above segmentation techniques can help to study thoroughly and semi-autonomously the surface of Mars. Such a thorough study will be essential in choosing landing sites for the PHOENIX lander in 2007, and for the Mars Science Laboratory rover in 2009.

BACKGROUND INFO:
http://www.cab.inta.es/~CYBORG/cyborg.summary.html