Google Maps Is Upgrading Street View and You Can Help

[Narrator] For more than a decade,

Google's street view cameras have been recording

the world around us.

Street view over the years has gone to literally,

every corner of the planet.

We've gone to dense urban areas.

We've gone to the tops of mountains.

We've taken it under water and we've taken it to

islands populated primarily by sheep.

We've taken it to the International Space Station.

[Narrator] Now there's a new camera rig on the road

that will capture our world in greater detail

than ever before.

On the top you'll see the rosette.

In the past the ball was about this big

and now we've went from 15 cameras down to seven cameras.

There are 20 megapixel sensors behind

each one of these lenses.

[Narrator] Expect shaper,

more contrast images and fewer stitching errors

like these from Santa Cruz, California,

one of the first places the new cameras have recorded.

[Steve] The two cans in the front and the back

are laser radars.

Those are used to position us in the world.

Then down here is the side looking HD cameras,

which are designed for picking up business names

and street signs.

[Narrator] That data is then analyzed by Google's

artificial intelligence,

which has learned to know what part of an image

warrant the most attention.

[Andrew] This is an actual example of a real

street view image taken in Brazil of a business.

A tire shop I actually think.

In addition to actually being able to pull out useful

information like what these words are,

it does something a little neat.

It realizes the title,

in this case I'm going to screw up the pronunciation

because I don't speak Portuguese but,

it's Zelina Pneus.

But it manages to actually ignore the brand names,

the phone numbers, stuff like that,

that's kind of spurius.

What's interesting about that is that's kind of what

a human is doing too.

One of the other nice things that the detection mechanism

allows you to do is,

this model is able to learn kind of in an end to end sense

what normalized text looks like.

In this particular situation,

it's learned that when it sees AV on a street sign,

what we typically care about is the word avenida.

P-R-E-S period typically refers to presidente

at least in this market.

[Narrator] This sort of image recognition applied to

mapping can not only make for more accurate directions,

but it can also help keep up with rapid growth and change

in mega cities like Lagos Nigeria.

[Andrew] The Lagos work was actually a really neat

application of a number of these pipelines.

We've built some of them deep learning base,

that allow us to look at a couple of different

inputs including street view imagery

and very quickly populate the map in areas where otherwise

we might have to spend an awful lot of protracted effort.

[Narrator] What Google is doing here

is what Google has always done,

and done so well.

It's indexing.

[Charles] People expect that we've indexed

the world's websites and generally the world's information.

Indexing the world's spaces is actually a far more

challenging problem even if people have the same

expectation of us.

So by being more able to deliver on those expectations,

by putting the power into people's hands,

we think that we can have happier users.

[Narrator] Now Google wants you the public to help

by contributing street views of your own.

Thanks to an explosion of relatively inexpensive

360 degree cameras,

nearly anyone can contribute their own fresh

images to Google Maps.

The newer cameras are now increasingly capable of

360 video capture at higher resolutions.

So with the street view app,

what we've done is really lean into that by

allowing you to collect video,

360 video on the go while you're driving,

biking, walking, whatever it may be and actually just

collect a ton of individual 360 photos from that.

Stringing that back together into the traditional street

view experience that you're familiar with,

from something like this.

[Narrator] One caveat,

Google won't automatically scrub user generated captures

to blur out faces but you can choose to do so in the app.

[Charles] Always make sure that you respect the privacy

tensions and requests of anybody that you're passing by.

A lot of people don't want to be in public imagery.

[Narrator] So record carefully.

Knowing that your contributions will help Google in its

quest to index the world and maintain its map's dominance.

[Jen] We see our job on the Google Maps team

as building a data set that helps really capture

all the knowledge understanding of the real world

and bring that into digital form.

A lot of what we're doing,

whether it's with street view with our user generated

content and ability for people to contribute their own

knowledge and understanding,

is really helping it build a deeper data set of knowledge

about the world so that when people come to Google and ask

questions about the world around them,

we have more and more of the answers.