Live data from Hacker News

How Google Builds Its Maps

theatlantic.com

111–113 of 113 posts

Re: How Google Builds Its Maps

#111
post #77

Earlier quoted context omitted.

> It's begging for a machine vision application, which records video and... A simpler use case for computer vision I could think of with Openstreetmap would be to automatically extract roads and building shapes from satellite imagery (though I hope that the license to use Bing would extend to something like this). > As an aside, are there any efforts afoot to decentralise OSM? Well, even Wikipedia's not decentralized…

Automatic tracing of roads from Bing provided aerial photos, with code provided by MSFT, 18 months ago: http://youtu.be/LR0WV2dGIRc http://magicshop.cloudapp.net/Default.aspx

Speaking as someone that wrote some code that interfaces with the Magicshop API, it works nowhere near as good as described in the video. It's a simple color matching algorithm that gets easily fooled by trees and dry grass.

The project is a great start, but has basically been abandoned.

Re: How Google Builds Its Maps

#112
post #101

Earlier quoted context omitted.

You can do that by clicking on "My Places".

If I am looking at an address and want directions to it clicking on 'My places' only shows me "my places", it does not give me any obvious way to get directions from the current address to one of "my places". Clicking on any of "my places" will change the map to view that place, causing the user to lose the address of the place they want to go to/from. Edit: I see that "my places" now has a "Home address" which isn't…

I absolutely agree that it needs a rethinking in terms of UI. Now that they are trying to consolidate google products, maps should be more aggressive in identifying useful destinations. I'm guessing they want to err on the side of not creeping people out, but I expect a popup introducing these features soon.

Re: How Google Builds Its Maps

#113

Earlier quoted context omitted.

Automatic tracing of roads from Bing provided aerial photos, with code provided by MSFT, 18 months ago: http://youtu.be/LR0WV2dGIRc http://magicshop.cloudapp.net/Default.aspx

Speaking as someone that wrote some code that interfaces with the Magicshop API, it works nowhere near as good as described in the video. It's a simple color matching algorithm that gets easily fooled by trees and dry grass. The project is a great start, but has basically been abandoned.

I think the more important point though, is that both the permission and imagery is there (and the imagery is very high quality) so there's various other projects that e.g. trace outlines of lakes from the imagery (again basic color matching) or identify tennis courts via computer vision etc.
Post reply on HN