ANYONE strolling by the Mars Society鈥檚 Desert Research Station near Hanksville, Utah, in February would have been baffled to see two men in spacesuits bouncing around while pointing cellphones at rocks.
No, they hadn鈥檛 taken leave of their senses. The pair were testing out an imaging algorithm designed to automatically pinpoint areas of geological interest on future crewed or robotic missions to the Red Planet (see image). These might include unusual rock formations or signs of organic matter that could indicate life. The algorithm should give astronauts the eyes of a trained human geologist 鈥� though, for now, it is fed its pictures via a regular Nokia camera-phone (see image).
Advertisement
To test the algorithm, a team led by Patrick McGuire from the University of Chicago used the phones to take a series of pictures of the Utah desert, an area often used to stand in for Mars鈥檚 dusty environment (see image). The photos were then transferred via a Bluetooth wireless link to a nearby laptop running the vision algorithm.
The software picks out unfamiliar areas in the photo by classifying the image according to its hue, intensity and brightness. It needs about five images for the system to 鈥渓earn鈥� the basic lie of the land and its normal colours 鈥� and after that it only highlights new, unexpected colours as interesting. McGuire says that in a future mission, this algorithm could be embedded into a rover鈥檚 on-board computer to let it select targets for exploration and analysis. Currently the algorithm works at a 鈥渃hild-like鈥� level of accuracy and takes about 2 minutes to process an image, but McGuire promises later versions will be 鈥渟uperhuman鈥�.
鈥淭he software classifies unfamiliar areas in the photo according to hue, intensity and brightness鈥�
However, Tanja Zegers from the University of Utrecht in the Netherlands doubts the system could ever perform as well as a human geologist. She points out that texture rather than colour is key. 鈥淕eological analysis is a very complex process, in which pattern recognition plays a large role. This only works if it is carried out by a trained brain.鈥� She says it would be better to transfer the pictures to a geologist back on Earth who can then analyse the site.
McGuire, who presented the latest results from the project at the European Space Agency鈥檚 Comparative Planetology conference in the Netherlands on 15 May, says the algorithm will soon be able to detect novel textures, porosities and rock strata formations.
But what of the snap-happy desert astronauts? 鈥淭hey had a little bit of a 鈥榖ig-finger鈥� problem with the mobile phonecams when they were suited up, but they could still complete some useful tests,鈥� McGuire says.
The Chicago team don鈥檛 expect phonecams to be used on the moon or Mars. Instead, a camera would be attached to a spacesuit or helmet 鈥� or in a space-certified camera device that works well with gloved fingers.
