欢迎点云相关产学研的学者和团体加入我们。
笔者评:从街景地图到下一代真正的三维地图还有多远,想象一下,全球人每天照的照片,积累下来通过立体匹配多目重建出来,能否将整个地球的表面三维重建出来,对的,现在Google就在做这个事情。当然,有很多研究是说用很多网上搜索到的照片,其实如果真正的实用了,肯定会有这样的数据提供商的。这个设想的基础就是基于点云获取技术中的一种,多目重建获取点云。华盛顿大学在这块做不很多基础研究,现在也在google下代地图上基于此在合作。如果对这些感兴趣,学者Yasutaka Furukawa的相关研究非常有趣。
SAN FRANCISCO—Google is trying to use algorithms to replicate what humans can do with a pair of revamped Google Maps features, the company said Friday. The result of that work is the delivery of hyper-localized maps that better reflect what real people find useful as they travel around, and the refinement of the 3D Tour engine in Google Maps.
These redesigned features in Google Maps aren't perfect but they're a good start, said Google Maps lead engineer Yatin Chawathe, who along with lead designer Jonah Jones offered folks a look at how the new Maps elements work here at Google I/O.
"These techniques are very much in their infancy. We've taken a big leap with the new Google Maps but we believe there are still a lot of very difficult computer science problems to solve that will make the Google Maps experience even better and more valuable," Chawathe said. Walking through the new design and underpinnings of Google's local area mapping, Jones said the idea was to give people the kind of important information about an area that a local might draw up on a cocktail napkin for a visitor (pictured). That meant bringing to the front street and place names that don't always turn up on a standard, zoomed-in Google Map while fading or even eliminating less useful information.
But how to replicate the kind of maps and directions a knowledgeable local would draw for a bewildered tourist? The answer, as is par for the course with Google, was to create smarter algorithms to crunch through the data generated by all the people using Google Maps to get around a particular locale, such as Pullens Gardens in London, the example used by Jones and Chawathe.
It also helps to have tons of user data, Chawathe noted.
"The key is the immense amount of information Google has access to," he said, referring to how Google was able to develop the new localized mapping feature and its expanded 3D Tours.
Now, when a Google Maps users wants information on a place like Pullens Gardens, the redesigned Maps engine "does a very localized search around the immediate vicinity and pulls out very small streets around the destination spot that wouldn't turn up on a general Google Map," Jones said. Street and place information that has proven less useful to most Google Maps users doesn't disappear, but it's dimmed out in gray. The idea isn't just to provide the most useful map information, however—the redesign is also an attempt to recapture with digital maps some of the magic of the more personalized paper maps we used to rely upon.
"There's something missing from the days when we had an analog map to get around while traveling. You would put annotations on top of the map. Those maps were drawn for you, unique for you. And they have a nostalgia value when you pull them out a few years later and look at them to reflect on your travels," Jones said.
Cinematography by the Numbers
Google's improved 3D Tours also use the wealth of data Google is able to access to algorithmically create an experience a human film director might produce, specifically using "a lot of techniques from cinematography to make our 3D Tours more immersive and smooth than flipping through pictures in a photo album" but still light enough from a The process for building a 3D Tour of the Notre Dame Cathedral, for example, begins with selecting the best publicly available photos out of thousands to capture
multiple angles, long-range shots of the structure, close-ups of important details, particularly "picturesque" photos, and interstitial images to create smooth transitions in the tour, Chawathe explained.
What follows is a remarkably exact process for taking all of those images and meticulously stitching them together.
"Complex algorithms take these photos and try to detect precisely where in the real world each picture was taken, and then they calculate how every single pixel in a picture corresponds to the pixels in all the other pictures," Chawathe said.
What you get is a 3D point cloud (pictured above right) that is used as the baseline geometry for the creation of an immersive, remarkably smooth visual tour of a landmark building, city center, or basically any other locale where enough photography exists to create one. "It's a really interesting challenge. We're trying to generate algorithmically with 3D Tours what a really talented human filmmaker could do. And they would do it much better, of course, but we see this as one step forward in the process of making better and better maps for people," Jones said.
参考:
http://www.pcmag.com