Name: 從街景到人工智能--谷歌地圖如何繪製世界地圖 (From Street View To AI — How Google Maps Mapped The World)
Uploaded: 2025-01-21T01:44:02.000Z
Duration: 15 min 19 s

分詞構句

Behind this unmarked garage door, pulling out onto the streets of Palo Alto, California, is a fleet of cars mounted with cameras, helping Google to map the world, something it's now been doing for 20 years.

在加利福尼亞州帕洛阿爾託的街道上，在這扇沒有標誌的車庫門後面，是一支安裝了攝像頭的車隊，幫助谷歌繪製世界地圖。

Here is where we're both repairing, building the cameras, keeping them maintained.

我們都在這裡修理、製造攝影機，並對它們進行維護。

Google's cameras have taken billions of images to map millions of miles in more than 100 countries, and they've come a long way since Street View images first published in 2007.

自 2007 年首次發佈街景影像以來，谷歌的相機已經拍攝了數十億張影像，繪製了 100 多個國家的數百萬英里地圖。

It's 500 pounds, and we needed a forklift to put it on top of the van.

它有 500 磅重，我們需要一輛叉車才能把它放到貨車頂部。

By 2008, the nine camera systems were 75 pounds, mounted on everything from snowmobiles to tricycles.

到 2008 年，這九套攝像系統重達 75 磅，安裝在從雪地車到三輪車等各種交通工具上。

是以，我們得以打開歐洲市場。

Europe has a lot of pedestrian areas and we can't drive through those.

歐洲有很多步行區，我們不能開車通過。

In 2010 came 15 camera 50-pound spheres, wearable by backpackers, underwater divers, even camels.

2010 年，15 臺重達 50 磅的球形相機問世，揹包客、水下潛水員甚至駱駝都可以佩戴。

And in 2018 came an even lighter 40-pound system I tried wearing on my back.

到了 2018 年，我又試著把一個更輕的 40 磅系統背在身上。

This is like what it's like to backpack with a baby.

We went to Grand Canyon in Machu Picchu, places that you can only really walk into.

我們去了大峽谷和馬丘比丘，這些地方你只能真正走進去。

With more than 2 billion monthly users, Google Maps is the world's top navigation app.

谷歌地圖擁有超過 20 億的月度用戶，是全球頂級導航應用程序。

And now Alphabet-owned Maps is using generative AI to keep ahead of the pack, with new consumer-facing features and AI analysis of billions of images to keep the map up to date.

現在，Alphabet 旗下的地圖公司正在利用人工智能生成技術保持領先地位，推出面向消費者的新功能，並對數十億張圖片進行人工智能分析，以不斷更新地圖。

We're making 100 million updates to the map every single day.

我們每天對地圖進行一億次更新。

We went to Silicon Valley for a ride-along with Google's mapping fleet and a first-ever on-camera look at the Street View Service Center, where we sat down with the head of Maps to talk traffic, data privacy and all the changes coming thanks to generative AI.

我們來到硅谷，與谷歌的地圖車隊進行了一次 "親密接觸"，並首次在鏡頭前參觀了街景服務中心，在那裡我們與地圖部門的負責人一起討論了交通、數據隱私以及生成式人工智能帶來的所有變化。

There was a time when paper maps were king.

Then it was MapQuest printouts and standalone GPS units from companies like Garmin and TomTom.

然後是 MapQuest 打印輸出以及 Garmin 和 TomTom 等公司的獨立 GPS 設備。

Then in 2005, Google made a bold move into Maps, following an idea from co-founder Larry Page.

2005 年，根據聯合創始人拉里-佩奇的想法，谷歌大膽進軍地圖領域。

我們就是從這裡開始繪製地圖的。

About 20 years ago, Larry drove down some of these streets with a video camera and handed it to someone and said, hey, what can you do with this?

大約 20 年前，拉里帶著一臺攝影機在這些街道上行駛，然後把攝影機遞給別人說，嘿，你能用這個做什麼？

Maria Biggs has been with Google for more than 10 years.

瑪麗亞-比格斯在谷歌工作了 10 多年。

She drove us around its headquarters in a car equipped with the latest generation Street View camera, first introduced in 2022.

她開著一輛配備了最新一代街景攝像頭的汽車帶我們參觀了公司總部，該攝像頭於 2022 年首次推出。

It's the first Street View camera that can be used on any car rather than being built in.

這是第一個可以在任何汽車上使用的街景攝像頭，而不是內置的。

As you can imagine, it's not tied to a specific car.

可以想象，它並不與特定的汽車掛鉤。

我們可以租一輛車，比如說，像一個小島一樣。

So we're going to Hawaii with these next-generation camera systems because we don't have to ship the whole car.

是以，我們將在夏威夷使用這些下一代攝像系統，因為我們不必整車運輸。

Street View cameras are a significant part of how Google gathers data for Maps, but it also relies on satellite and aerial images, 1,000-plus third-party sources and info from local governments and directly from Maps users.

街景相機是谷歌為地圖收集數據的一個重要部分，但它也依賴於衛星和航空影像、1000 多個第三方來源以及來自地方政府和直接來自地圖用戶的資訊。

All this allows it to offer Maps in a whopping 250-plus countries and territories.

所有這一切使其能夠在多達 250 多個國家和地區提供地圖服務。

Now, the more nimble cameras are helping Google make updates to dozens of those countries and map at least three new ones, Bosnia and Herzegovina, Namibia and Liechtenstein.

現在，更加靈活的攝像頭正在幫助谷歌對其中數十個國家進行更新，並繪製至少三個新國家的地圖，即波斯尼亞和黑塞哥維那、那密比亞和列支敦士登。

We can easily just take it to the next island without having to ship the entire car.

我們可以輕鬆地把它帶到下一個島嶼，而不必把整輛車運到那裡。

And so some places that we haven't refreshed, collected in, say, 10 years, we're going to be able to easily move these cameras around.

是以，在一些我們已經 10 年沒有更新和收集的地方，我們將能夠輕鬆地移動這些攝影機。

It used to be an extremely manual pipeline for any digital mapmaker.

對於任何數字地圖製作者來說，這曾經是一個極其繁瑣的手工操作流程。

But increasingly, you have these various IoT devices, this combination of sensing, processing and connectivity that allows you to update this map more automatically and therefore potentially more cheaply and in a way that is ultimately more scalable.

但是，越來越多的各種物聯網設備，將傳感、處理和連接結合在一起，讓你可以更自動地更新地圖，是以可能會更便宜，而且最終更具可擴展性。

The way it works now, Bigg says, is the cameras record images locally as they're driven around by a fleet of Street View employees who ingest the images once they return to a hub, like this Street View garage in Palo Alto.

比格說，現在的工作方式是，由街景公司員工組成的車隊將攝影機開到各地時在在地錄製影像，而一旦攝影機返回中心（比如帕洛阿爾託的這個街景車庫），街景公司員工就會攝取影像。

How many other garages like this are there around the world?

So we have three total, one in each region, one in APAC, one in Europe and one here.

是以，我們總共有三個，每個地區一個，亞太地區一個，歐洲一個，這裡一個。

Also on site here is the Street View service center.

This is where all of the cameras come through when they move from different locations.

當攝影機從不同地點移動時，所有攝影機都會從這裡經過。

This is a next-gen camera system where it has been to Australia.

這是一個新一代的攝像系統，曾在澳洲使用過。

We did some really extreme testing in really hot tropical areas.

我們在非常炎熱的熱帶地區進行了一些非常極端的測試。

And so we bring it back and we make sure everything's still functional.

是以，我們把它運回來，確保一切功能正常。

We clean it up before it goes back out to another country.

在運回另一個國家之前，我們將其清理乾淨。

More images and more detail from the new cameras also provide a rich data set for Google's AI models, resulting in a flurry of new consumer-facing features in the last couple of years.

新相機拍攝的更多影像和更多細節也為谷歌的人工智能模型提供了豐富的數據集，從而在過去幾年中推出了大量面向消費者的新功能。

Think of it like a supercharged version of Street View, using AI to fuse together billions of Street View and aerial images to show a lifelike preview of each step on the route.

可以把它想象成 "街景 "的升級版，利用人工智能將數十億張 "街景 "和航拍影像融合在一起，栩栩如生地預覽路線上的每一步。

We use the imagery and neural radiance fields, which is a generative AI technique, to actually take a photo and transform it into a 3D representation of a place.

我們使用影像和神經輻射場（一種生成式人工智能技術）來實際拍攝一張照片，並將其轉化為一個地方的三維呈現。

So almost like you can look into a restaurant.

是以，你幾乎可以看到一家餐廳。

And we overlay the conditions, like what is the actual, how busy it is and what the weather might be, air quality.

我們會疊加各種條件，如實際情況、繁忙程度、天氣和空氣質量。

And in October, Google announced a bunch more AI features.

今年 10 月，谷歌又發佈了一系列人工智能功能。

One major one, Maps, is now enabled with Gemini, Google's generative AI chatbot.

其中最重要的一項功能--地圖，現在已經啟用了谷歌的生成式人工智能哈拉機器人 Gemini。

Where could I watch Sunday's football game while drinking a beer and bring my dog?

我可以在哪裡邊喝啤酒邊看週日的橄欖球賽，還能帶著我的狗一起去？

This is historically kind of hard information to find.

這在歷史上是很難找到的資訊。

And now, with a quick tap, you can learn about it.

現在，只需輕輕一點，您就可以瞭解它。

Chris Phillips joined Google in 2021 to head up all of Geo, the division in charge of Maps.

克里斯-菲利普斯於 2021 年加入谷歌，負責管理整個地理部門，即地圖部門。

He showed us some of the newest AI features.

他向我們展示了一些最新的人工智能功能。

This place has 2,000 reviews, kind of a lot of reading to do.

這個地方有 2,000 條評論，要讀的東西還真不少。

We summarize it using Gemini for you right away.

Gemini is also enabling voice-activated reports in Waze, which Google bought in 2013 for $1.3 billion.

Gemini 還在 Waze 中啟用了聲控報告功能，谷歌在 2013 年以 13 億美元收購了 Waze。

To get consumers to actually use these things, they need to be easy to use.

為了讓消費者真正使用這些東西，它們必須易於使用。

And a voice interface is the best kind of interface that you can have in the cabin.

語音界面是車廂內最好的界面。

It allows you to keep your eyes out on the road.

It allows you to keep your hands on the wheel.

它可以讓你的雙手緊握方向盤。

In Google Earth, Gemini is enabling custom visualizations that can help with high-level city planning tasks.

在谷歌地球中，雙子座可實現定製可視化，幫助完成高級城市規劃任務。

They might want to be thinking about, where should we put EV charging stations in this area?

他們可能需要考慮的是，我們應該在這個區域的什麼地方設置電動汽車充電站？

And we can help them map quickly with looking at, where's the current EV charging?

我們可以幫助他們快速繪製地圖，查看當前的電動汽車充電點在哪裡？

They can overlay information around population and quickly do the surveying work and help them do that to make those decisions faster.

他們可以在人口周圍疊加資訊，快速完成勘測工作，幫助他們更快地做出決策。

On public transit, there are now delay reports, alternate routes and details like subway entrance locations.

在公共交通方面，現在有延誤報告、備用路線和地鐵入口位置等詳細信息。

And in navigation, AI is enabling real-time reports of weather disruptions like unplowed roads, flooded areas and spots with low visibility.

在導航方面，人工智能能夠實時報告天氣干擾，如未鋪設道路、洪水氾濫地區和能見度低的地點。

And then we have this new arrival experience where when you get to your destination, we can help you with not just where to park, but even that last step of walking along the way.

此外，我們還提供了全新的到達體驗，當您到達目的地時，我們不僅能幫您找到停車位，甚至還能幫您走完最後一步。

While AI helps curate these details into an updated experience, it's being fed real-time data that often comes from you, the users.

雖然人工智能有助於將這些細節整理成最新的體驗，但它所獲得的實時數據往往來自於用戶。

We have over 500 million people who are contributing updates to the map every day.

每天都有超過 5 億人對地圖進行更新。

That's things like reviews or keeping the hours up to date and even like what to expect as far as issues on the road on the way there.

比如評論或及時更新時間，甚至是在路上遇到的問題。

But users also provide data to Google Maps passively, too.

但用戶也會被動地向谷歌地圖提供數據。

Traffic delays, for instance, are predicted in part using live information like speed and drive time from the millions of drivers who use maps each day.

例如，交通延誤的預測部分是利用每天使用地圖的數百萬駕駛員提供的實時資訊，如車速和駕駛時間。

So what was it like on this roadway on this day of the week at this time of day?

那麼，在一週中的這一天，一天中的這個時間段，這條道路上的情況如何呢？

And we use anonymized data so it's not linked to any individual user.

我們使用的是匿名數據，是以不會與任何個人用戶相關聯。

So we aggregate that information to know the historical trend.

是以，我們彙總了這些資訊，以瞭解歷史趨勢。

Data ownership and data privacy have always been one of the major perception barriers that Google has had to overcome.

數據所有權和數據隱私一直是谷歌必須克服的主要認知障礙之一。

Over the years, Google has put in safeguards to help.

多年來，谷歌已經採取了一些保障措施。

We built technology to blur faces and even blur things like license plates.

我們開發的技術可以模糊人臉，甚至模糊車牌等。

Location history is valuable data for Google, but users can now turn off location history or delete places they've been.

位置歷史記錄對谷歌來說是非常寶貴的數據，但現在用戶可以關閉位置歷史記錄或刪除他們去過的地方。

And some, like abortion clinics or domestic violence shelters, are auto-deleted.

還有一些，如墮胎診所或家庭暴力庇護所，會被自動刪除。

And in a privacy-conscious move in December, Google started keeping location history on device instead of in the cloud tied to users' Google accounts, making it tougher for authorities to access location history.

12 月，谷歌為了保護用戶隱私，開始將定位歷史記錄保存在設備上，而不是與用戶的谷歌賬戶綁定在雲端，從而使當局更難獲取定位歷史記錄。

And users can request blurring an area on Street View, say, to prevent thieves from analyzing their property.

用戶還可以要求模糊街景中的某個區域，比如說，防止小偷分析他們的財產。

But there is an upside to all this data, too.

不過，這些數據也有一個好處。

It gives Google a lot of power to warn drivers of hazards.

它賦予了谷歌很大的權力，可以向駕駛員發出危險警告。

Especially when you have conditions happening like a flooded road and we can surface it for other drivers in almost real time.

尤其是在發生水淹路面等情況時，我們幾乎可以實時為其他駕駛員提供資訊。

It really does make a difference in helping avoid what could be other problems happening on the road.

這確實有助於避免路上可能發生的其他問題。

Google Maps gets updated with data from Waze, where reporting is core to the experience for loyal users.

谷歌地圖會根據 Waze 的數據進行更新，而報告是忠實用戶體驗的核心。

Built into the DNA of Waze is this idea of two-way interaction and what has been a very virtuous cycle for Waze and therefore in directing for Google.

Waze 的基因中就包含了這種雙向互動的理念，這對 Waze 來說是一個非常良性的循環，是以對谷歌來說也是如此。

字幕列表影片播放

從街景到人工智能--谷歌地圖如何繪製世界地圖 (From Street View To AI — How Google Maps Mapped The World)

specific

access

opportunity

experience