ABOUT THE SPEAKER
Michael Rubinstein - Research scientist, Google
Computer scientist Michael Rubinstein and his team have developed a "motion microscope" that can show video footage of barely perceivable movements, like breaths and heartbeats.

Why you should listen

Michael Rubinstein zooms in on what we can't see and mangnifies it by thirty or a hundred times. His "motion microscope," developed at MIT with Microsoft and Quanta Research, picks up on subtle motion and color changes in videos and blows them up for the naked eye to see. The result: fun, cool, creepy videos.

Rubinstein is a research scientist at a new Cambridge-based Google lab for computer vision research. He has a PhD in computer science and electrical engineering from MIT.

More profile about the speaker
Michael Rubinstein | Speaker | TED.com
TEDxBeaconStreet

Michael Rubinstein: See invisible motion, hear silent sounds

迈克尔·鲁宾斯坦: 看到不可见的运动,听到沉默的声音。酷?令人毛骨悚然?我们不能决定

Filmed:
2,075,056 views

看看“运动显微镜”,一个视频处理软件,可以把运动或颜色的微小变化转化为肉眼可见。视频研究员迈克尔·鲁宾斯坦播放了一段令人惊异的视频,视频中展示了这项技术可以从一段视频获取某人的脉搏和心跳。看他通过放大声音在薯片袋上造成的震动,重构出声音。这项技术令人叹为观止的邪恶应用,您有必要眼见为实。
- Research scientist, Google
Computer scientist Michael Rubinstein and his team have developed a "motion microscope" that can show video footage of barely perceivable movements, like breaths and heartbeats. Full bio

Double-click the English transcript below to play the video.

00:13
So over the past过去 few少数 centuries百年,
microscopes显微镜 have revolutionized革命性 our world世界.
0
1114
7492
过去的几个世纪,显微镜改变了世界。
00:21
They revealed透露 to us a tiny world世界
of objects对象, life and structures结构
1
9036
5216
它揭示了肉眼难以看到的
00:26
that are too small for us
to see with our naked eyes眼睛.
2
14252
2906
物体,生命和结构的微观世界。
00:29
They are a tremendous巨大 contribution贡献
to science科学 and technology技术.
3
17158
3019
它对科技进步做出了巨大的贡献。
00:32
Today今天 I'd like to introduce介绍 you
to a new type类型 of microscope显微镜,
4
20177
3227
今天我要向您介绍一类新型显微镜,
00:35
a microscope显微镜 for changes变化.
5
23404
2578
观察变化的显微镜。
00:37
It doesn't use optics光学
like a regular定期 microscope显微镜
6
25982
2902
它不是通常所说的光学显微镜
00:40
to make small objects对象 bigger,
7
28884
1997
可以放大微小物体,
00:42
but instead代替 it uses使用 a video视频 camera相机
and image图片 processing处理
8
30881
4376
它利用视频摄像机和图像处理
00:47
to reveal揭示 to us the tiniest最小的 motions运动
and color颜色 changes变化 in objects对象 and people,
9
35257
5256
来揭示物体和人的微小的颜色和运动变化,
00:52
changes变化 that are impossible不可能
for us to see with our naked eyes眼睛.
10
40513
3842
这些变化是光凭肉眼难以分辨的。
00:56
And it lets让我们 us look at our world世界
in a completely全然 new way.
11
44355
4120
它可以让我们以全新角度看世界。
01:00
So what do I mean by color颜色 changes变化?
12
48475
1910
我所说的颜色变化是什么意思呢?
01:02
Our skin皮肤, for example,
changes变化 its color颜色 very slightly
13
50385
2832
举例来说,我们的皮肤,当血液流动时
01:05
when the blood血液 flows流动 under it.
14
53217
1997
皮肤表面的颜色有微弱变化。
01:07
That change更改 is incredibly令人难以置信 subtle微妙,
15
55214
2397
变化十分微小,
01:09
which哪一个 is why, when you
look at other people,
16
57611
2063
小到当你观察坐在
01:11
when you look at the person
sitting坐在 next下一个 to you,
17
59674
2251
你身边的人的时候,
01:13
you don't see their skin皮肤
or their face面对 changing改变 color颜色.
18
61925
3575
你不能看到他脸上皮肤的颜色变化。
01:17
When we look at this video视频 of Steve史蒂夫 here,
it appears出现 to us like a static静态的 picture图片,
19
65500
4360
我们看看史蒂夫的这段视频,看上去好像是静态图片,
01:21
but once一旦 we look at this video视频
through通过 our new, special特别 microscope显微镜,
20
69860
3860
但是一旦我们通过我们新的、特别的显微镜看这段视频,
01:25
suddenly突然 we see
a completely全然 different不同 image图片.
21
73720
2600
我们会看到完全不同的影像。
01:28
What you see here are small changes变化
in the color颜色 of Steve's史蒂夫的 skin皮肤,
22
76320
3930
您在这里看到的是史蒂夫皮肤颜色微小变化
01:32
magnified放大 100 times
so that they become成为 visible可见.
23
80250
4436
放大100倍之后的肉眼可见的效果。
01:36
We can actually其实 see a human人的 pulse脉冲.
24
84686
3267
我们可以明确的看到人体的脉搏。
01:39
We can see how fast快速
Steve's史蒂夫的 heart is beating跳动,
25
87953
3227
我们可以看到史蒂夫心脏跳动得多快,
01:43
but we can also see the actual实际 way
that the blood血液 flows流动 in his face面对.
26
91180
5355
我们还可以明确地看到血液是如何流经史蒂夫脸部的。
01:48
And we can do that not just
to visualize想象 the pulse脉冲,
27
96544
2631
我们不仅可以看到脉搏,
01:51
but also to actually其实
recover恢复 our heart rates利率,
28
99175
3471
还可以推测心跳速率,
01:54
and measure测量 our heart rates利率.
29
102646
1793
得到我们心跳速率。
01:56
And we can do it with regular定期 cameras相机
and without touching接触 the patients耐心.
30
104439
4453
我们可以通过常用摄像机,不需碰触病人就可以实现。
02:00
So here you see the pulse脉冲 and heart rate
we extracted提取 from a neonatal新生儿 baby宝宝
31
108892
5617
这里您可以看到我们提取的新生儿宝宝的脉搏和心跳速率
02:06
from a video视频 we took
with a regular定期 DSLRDSLR camera相机,
32
114509
2881
这是从常规数码单反相机视频中提取的,
02:09
and the heart rate measurement测量 we get
33
117390
1816
我们测出的心率
02:11
is as accurate准确 as the one you'd get
with a standard标准 monitor监控 in a hospital醫院.
34
119206
4811
和医院监测器测出的一样精确。
02:16
And it doesn't even have to be
a video视频 we recorded记录.
35
124017
2642
而且我们的软件甚至不需要我们自己专门拍摄的视频。
02:18
We can do it essentially实质上
with other videos视频 as well.
36
126659
2995
我们可以对其他视频作同样处理。
02:21
So I just took a short clip
from "Batman蝙蝠侠 Begins开始" here
37
129654
3901
这是我截取《蝙蝠侠:侠影之谜》(Batman Begins)的一小段片断
02:25
just to show显示 Christian基督教 Bale's贝尔的 pulse脉冲.
38
133555
1904
显示演员克里斯蒂安·贝尔(Christian Bale)的脉搏。
02:27
(Laughter笑声)
39
135459
1822
(笑声)
02:29
And you know, presumably想必
he's wearing穿着 makeup化妆,
40
137281
2123
你知道,他应该是化了妆了,
02:31
the lighting灯光 here is kind of challenging具有挑战性的,
41
139404
1953
灯光也是挑战,
02:33
but still, just from the video视频,
we're able能够 to extract提取 his pulse脉冲
42
141357
2951
但是我们同样从这段视频中提取到了他的脉搏
02:36
and show显示 it quite相当 well.
43
144308
2018
演示效果不错。
02:38
So how do we do all that?
44
146326
1920
我们是如何做到的呢?
02:40
We basically基本上 analyze分析 the changes变化
in the light that are recorded记录
45
148246
4598
首先我们分析了录像中每个像素
02:44
at every一切 pixel像素 in the video视频 over time,
46
152844
2271
的光线随时间发生的变化,
02:47
and then we crank曲柄 up those changes变化.
47
155115
1798
我们放大这些变化。
02:48
We make them bigger
so that we can see them.
48
156913
2162
经过放大使我们肉眼可以看见这些变化。
02:51
The tricky狡猾 part部分 is that those signals信号,
49
159075
1902
棘手的部分是,那些信号,
02:52
those changes变化 that we're after,
are extremely非常 subtle微妙,
50
160977
2933
那些我们要放大的信号十分微小,
02:55
so we have to be very careful小心
when you try to separate分离 them
51
163910
2779
所以我们必须把它们和视频中存在的噪音
02:58
from noise噪声 that always exists存在 in videos视频.
52
166689
3831
区分出来。
03:02
So we use some clever聪明
image图片 processing处理 techniques技术
53
170520
2995
所以我们应用了一些聪明的影像处理技巧
03:05
to get a very accurate准确 measurement测量
of the color颜色 at each pixel像素 in the video视频,
54
173515
3994
精确的测量每个像素中的颜色,
03:09
and then the way the color颜色
changes变化 over time,
55
177509
2670
然后得出颜色随时间的变化,
03:12
and then we amplify放大 those changes变化.
56
180179
2693
再然后我们放大这些变化。
03:14
We make them bigger to create创建 those types类型
of enhanced增强 videos视频, or magnified放大 videos视频,
57
182872
3980
我们可以放大这些变化,增强视频、或放大视频,
03:18
that actually其实 show显示 us those changes变化.
58
186852
2172
这样就可以演示出变化。
03:21
But it turns out we can do that
not just to show显示 tiny changes变化 in color颜色,
59
189024
4238
事实上,我们不仅可以展示颜色的细小变化,
03:25
but also tiny motions运动,
60
193262
2241
同样也可以展示微小的运动,
03:27
and that's because the light
that gets得到 recorded记录 in our cameras相机
61
195503
3576
这是因为我们的摄像机
03:31
will change更改 not only if the color颜色
of the object目的 changes变化,
62
199079
2810
不仅记录物体的颜色变化,
03:33
but also if the object目的 moves移动.
63
201889
2368
同样也记录了物体的运动。
03:36
So this is my daughter女儿
when she was about two months个月 old.
64
204257
3636
这是我女儿,那时她才两个月大。
03:39
It's a video视频 I recorded记录
about three years年份 ago.
65
207893
2999
这是我三年前拍的。
03:42
And as new parents父母, we all want
to make sure our babies婴儿 are healthy健康,
66
210892
3208
作为初为人父母,我们要确保孩子的健康,
03:46
that they're breathing呼吸,
that they're alive, of course课程.
67
214100
2542
当然是,确保呼吸,确保活着。
03:48
So I too got one of those baby宝宝 monitors显示器
68
216642
2142
所以我也买了个婴儿监控器
03:50
so that I could see my daughter女儿
when she was asleep睡着.
69
218784
2469
这也我可以在她睡着时看到她。
03:53
And this is pretty漂亮 much what you'll你会 see
with a standard标准 baby宝宝 monitor监控.
70
221253
3527
这是标准婴儿监控器所看到的。
03:56
You can see the baby's宝宝 sleeping睡眠, but
there's not too much information信息 there.
71
224780
3682
您可以看到孩子睡着了,但没有其他信息了。
04:00
There's not too much we can see.
72
228474
1604
我们看不到更多东西。
04:02
Wouldn't岂不 it be better,
or more informative信息, or more useful有用,
73
230078
2824
可不可以更好点儿,有更多信息,更有用,
04:04
if instead代替 we could look
at the view视图 like this.
74
232902
2990
就像这样。
04:07
So here I took the motions运动
and I magnified放大 them 30 times,
75
235892
6356
这里我放大了运动30倍,
04:14
and then I could clearly明确地 see that my
daughter女儿 was indeed确实 alive and breathing呼吸.
76
242248
3826
这样我就可以真切地看到我女儿还活着,在呼吸。
04:18
(Laughter笑声)
77
246074
2253
(笑声)
04:20
Here is a side-by-side并排侧 comparison对照.
78
248327
1922
这是效果对比。
04:22
So again, in the source资源 video视频,
in the original原版的 video视频,
79
250249
2483
同样,在原始视频上,
04:24
there's not too much we can see,
80
252732
1636
我们看不到什么,
04:26
but once一旦 we magnify放大 the motions运动,
the breathing呼吸 becomes much more visible可见.
81
254368
3707
但是一旦我放大了运动,呼吸变得肉眼可见了 。
04:30
And it turns out, there's
a lot of phenomena现象
82
258075
2070
事实上,我们用我们的“运动显微镜”
04:32
we can reveal揭示 and magnify放大
with our new motion运动 microscope显微镜.
83
260145
3623
可以揭示很多现象。
04:35
We can see how our veins and arteries动脉
are pulsing脉动 in our bodies身体.
84
263768
4564
我们可以看到体内静脉和动脉的脉动。
04:40
We can see that our eyes眼睛
are constantly经常 moving移动
85
268332
2628
我们可以看到我们的眼睛在不断运动
04:42
in this wobbly摇摆不定 motion运动.
86
270960
1887
在晃动中。
04:44
And that's actually其实 my eye,
87
272847
1509
这实际上是我的眼睛,
04:46
and again this video视频 was taken采取
right after my daughter女儿 was born天生,
88
274356
3065
同样,这段视频拍摄于我女儿出生不久之后,
04:49
so you can see I wasn't getting得到
too much sleep睡觉. (Laughter笑声)
89
277421
4202
所以你可以看到我睡得不多。(血丝)(笑声)
04:53
Even when a person is sitting坐在 still,
90
281623
2716
即使当人在静止坐着,
04:56
there's a lot of information信息
we can extract提取
91
284339
2044
我们也可以提取许多信息
04:58
about their breathing呼吸 patterns模式,
small facial面部 expressions表达式.
92
286383
3529
如呼吸节律,面部表情微小变化。
05:01
Maybe we could use those motions运动
93
289912
1625
也许这可以
05:03
to tell us something about
our thoughts思念 or our emotions情绪.
94
291537
3154
告诉我们,我们的想法或情绪。
05:06
We can also magnify放大 small
mechanical机械 movements运动,
95
294691
3255
我们同样也可以方法微小的机械运动,
05:09
like vibrations振动 in engines引擎,
96
297946
1555
如引擎的震动,
05:11
that can help engineers工程师 detect检测
and diagnose诊断 machinery机械 problems问题,
97
299501
3692
这可以帮助工程师诊断机械问题,
05:15
or see how our buildings房屋 and structures结构
sway摇摆 in the wind and react应对 to forces军队.
98
303193
4738
或者看到建筑或结构的随风摇动。
05:19
Those are all things that our society社会
knows知道 how to measure测量 in various各个 ways方法,
99
307931
4581
我们知道这些变化可以通过其他方法测量,
05:24
but measuring测量 those motions运动 is one thing,
100
312512
2453
但是测量是一回事,
05:26
and actually其实 seeing眼看 those
motions运动 as they happen发生
101
314965
2276
肉眼看到的又是
05:29
is a whole整个 different不同 thing.
102
317241
2554
另一回事。
05:31
And ever since以来 we discovered发现
this new technology技术,
103
319795
3041
自从我们开发者这项新技术,
05:34
we made制作 our code available可得到 online线上 so that
others其他 could use and experiment实验 with it.
104
322836
3953
我们便把代码放到了网上,这样其他人可以使用试验它。
05:38
It's very simple简单 to use.
105
326789
1875
用起来很简单。
05:40
It can work on your own拥有 videos视频.
106
328664
2044
可以处理你自己拍的视频。
05:42
Our collaborators合作者 at Quanta广达 Research研究
even created创建 this nice不错 website网站
107
330708
3193
我们的合作伙伴,量研科技(Quanta Research),甚至建了个网站
05:45
where you can upload上载 your videos视频
and process处理 them online线上,
108
333901
2678
这里可以上传视频并在线处理,
05:48
so even if you don't have any experience经验
in computer电脑 science科学 or programming程序设计,
109
336579
3816
这样即使你没有计算机科学或编程经验
05:52
you can still very easily容易 experiment实验
with this new microscope显微镜.
110
340395
2936
也可以试验这种新型“显微镜”。
05:55
And I'd like to show显示 you
just a couple一对 of examples例子
111
343331
2404
我要演示几个其他的例子
05:57
of what others其他 have doneDONE with it.
112
345735
2735
别人用它做了什么。
06:00
So this video视频 was made制作 by
a YouTubeYouTube的 user用户 called TamezTamez85.
113
348470
5317
这段视频是一个叫Tamez85的YouTube用户作的
06:05
I don't know who that user用户 is,
114
353787
1463
我不知道他是谁,
06:07
but he, or she, used our code
115
355250
2345
但是他(她)用我们的代码
06:09
to magnify放大 small belly肚皮
movements运动 during pregnancy怀孕.
116
357595
3715
方法了孕妇腹部的运动。
06:13
It's kind of creepy爬行.
117
361310
1602
令人毛骨悚然。
06:14
(Laughter笑声)
118
362912
1613
(笑声)
06:16
People have used it to magnify放大
pulsing脉动 veins in their hands.
119
364525
4961
人们用它放在自己的腕部脉搏。
06:21
And you know it's not real真实 science科学
unless除非 you use guinea几内亚 pigs,
120
369486
3782
你知道要能被称为科学必须用到豚鼠,
06:25
and apparently显然地 this guinea几内亚 pig
is called Tiffany蒂芙尼,
121
373268
3390
显然这只豚鼠叫蒂芙妮,
06:28
and this YouTubeYouTube的 user用户 claims索赔
it is the first rodent啮齿动物 on Earth地球
122
376658
2949
这位YouTube用户声称
06:31
that was motion-magnified运动放大.
123
379607
2688
这是地球上首个运动被放大的啮齿类动物。
06:34
You can also do some art艺术 with it.
124
382295
2188
你也可以用它进行艺术创作。
06:36
So this video视频 was sent发送 to me
by a design设计 student学生 at Yale耶鲁.
125
384483
3018
这是耶鲁大学设计系学生发给我的视频。
06:39
She wanted to see
if there's any difference区别
126
387501
2137
她想看看
06:41
in the way her classmates同学 move移动.
127
389638
1522
她同学运动的方式有何不同。
06:43
She made制作 them all stand still,
and then magnified放大 their motions运动.
128
391160
4209
她让他们静止站立然后放大他们的运动。
06:47
It's like seeing眼看
still pictures图片 come to life.
129
395369
3378
者看上去像是图片有了生命。
06:50
And the nice不错 thing with
all those examples例子
130
398747
2433
这些例子的有意思的地方是
06:53
is that we had nothing to do with them.
131
401180
2296
我们没有进行干预。
06:55
We just provided提供 this new tool工具,
a new way to look at the world世界,
132
403476
3854
我们只是提供了新工具,一种看世界的新方法,
06:59
and then people find other interesting有趣,
new and creative创作的 ways方法 of using运用 it.
133
407330
5132
然后人们就找到其他有意思、新的创造性地方法使用这个工具。
07:04
But we didn't stop there.
134
412462
1764
我们不满足于此。
07:06
This tool工具 not only allows允许 us
to look at the world世界 in a new way,
135
414226
3251
这个工具不仅使我们有了看世界的新方法,
07:09
it also redefines重新定义 what we can do
136
417477
2368
同样它还重新定义了
07:11
and pushes the limits范围 of what
we can do with our cameras相机.
137
419845
3181
摄像机的功能限制。
07:15
So as scientists科学家们, we started开始 wondering想知道,
138
423026
2229
作为科学家,我们开始思考,
07:17
what other types类型 of physical物理 phenomena现象
produce生产 tiny motions运动
139
425255
3785
其他物理现象造成的微小振动
07:21
that we could now use
our cameras相机 to measure测量?
140
429040
2903
现在我们可以用摄像机来测量?
07:23
And one such这样 phenomenon现象
that we focused重点 on recently最近 is sound声音.
141
431943
4001
其中一种现象就是,声音。
07:27
Sound声音, as we all know,
is basically基本上 changes变化
142
435944
2105
声音,我们知道
07:30
in air空气 pressure压力 that
travel旅行 through通过 the air空气.
143
438049
2183
声音是一种压力波,依靠空气的压缩变化在空气中传播。
07:32
Those pressure压力 waves波浪 hit击中 objects对象
and they create创建 small vibrations振动 in them,
144
440232
3621
压力波碰到物体,就会引起物体本身的微小震动。
07:35
which哪一个 is how we hear
and how we record记录 sound声音.
145
443853
2532
这就是我们听到了录音的原理。
07:38
But it turns out that sound声音
also produces产生 visual视觉 motions运动.
146
446385
3668
这样,声音也会造成物体的视觉运动。
07:42
Those are motions运动
that are not visible可见 to us
147
450053
2833
这些运动,肉眼难于分辨
07:44
but are visible可见 to a camera相机
with the right processing处理.
148
452886
3001
但是通过处理摄像机却可见。
07:47
So here are two examples例子.
149
455887
1573
这里有两个例子。
07:49
This is me demonstrating示范
my great singing唱歌 skills技能.
150
457460
2614
这里我展示我伟大的歌唱技巧。
07:53
(Singing唱歌)
151
461064
1634
(歌唱)
07:54
(Laughter笑声)
152
462698
1436
(笑声)
07:56
And I took a high-speed高速 video视频
of my throat while I was humming低唱.
153
464134
2986
我哼唱时,我录了一段高速视频。
07:59
Again, if you stare at that video视频,
154
467120
1764
如果您盯着视频看的话,
08:00
there's not too much
you'll你会 be able能够 to see,
155
468884
2076
仅凭肉眼您是看不出什么端倪的,
08:02
but once一旦 we magnify放大 the motions运动 100 times,
we can see all the motions运动 and ripples涟漪
156
470960
4332
但是当我把运动方法100倍后我们就可以看到颈部的运动和波纹
08:07
in the neck颈部 that are involved参与
in producing生产 the sound声音.
157
475292
3274
这是由于声音震动造成的。
08:10
That signal信号 is there in that video视频.
158
478566
2740
信号就藏在这段视频中。
08:13
We also know that singers歌手
can break打破 a wine红酒 glass玻璃
159
481306
2670
我们知道,歌唱家可以震碎酒杯
08:15
if they hit击中 the correct正确 note注意.
160
483976
1463
如果他们发出正确的音符。
08:17
So here, we're going to play a note注意
161
485439
1765
我们播放一个音符
08:19
that's in the resonance谐振
frequency频率 of that glass玻璃
162
487204
2526
玻璃杯旁边音箱发出
08:21
through通过 a loudspeaker喇叭 that's next下一个 to it.
163
489730
2048
的共振频率。
08:23
Once一旦 we play that note注意
and magnify放大 the motions运动 250 times,
164
491778
4419
我们播放音符,然后方法运动250倍,
08:28
we can very clearly明确地 see
how the glass玻璃 vibrates振动
165
496197
2338
我们可以清楚的看到
08:30
and resonates共振 in response响应 to the sound声音.
166
498535
3570
玻璃杯是如何响应声音共振的振动的。
08:34
It's not something you're used
to seeing眼看 every一切 day.
167
502105
2420
这是我们日常见不到的。
08:36
But this made制作 us think.
It gave us this crazy idea理念.
168
504525
3529
但是这激发了我们思考。给我们一个疯狂的主意。
08:40
Can we actually其实 invert倒置 this process处理
and recover恢复 sound声音 from video视频
169
508054
5608
我们能不能通过分析声音在物体上造成的微小振动
08:45
by analyzing分析 the tiny vibrations振动
that sound声音 waves波浪 create创建 in objects对象,
170
513662
4035
从视频中逆向重构出声音来呢,
08:49
and essentially实质上 convert兑换 those
back into the sounds声音 that produced生成 them.
171
517697
4777
重构出原来的声音呢?
08:54
In this way, we can turn
everyday每天 objects对象 into microphones麦克风.
172
522474
4457
依靠这种方法,我们可以把任何物体变成麦克风。
08:58
So that's exactly究竟 what we did.
173
526931
2232
我们真的照做了。
09:01
So here's这里的 an empty bag of chips芯片
that was lying说谎 on a table,
174
529163
2816
这是一个空薯片袋,放在桌上,
09:03
and we're going to turn that
bag of chips芯片 into a microphone麦克风
175
531979
2825
我们要把这个薯片袋变为一个麦克风
09:06
by filming拍戏 it with a video视频 camera相机
176
534804
1591
用摄像机拍摄视频
09:08
and analyzing分析 the tiny motions运动
that sound声音 waves波浪 create创建 in it.
177
536395
3228
然后分析视频中声音造成的微小振动。
09:11
So here's这里的 the sound声音
that we played发挥 in the room房间.
178
539623
2796
这是我们在房间中播放的声音。
09:14
(Music音乐: "Mary玛丽 Had a Little Lamb羊肉")
179
542419
4434
(音乐:“Mary Had a Little Lamb”)
09:22
And this is a high-speed高速 video视频
we recorded记录 of that bag of chips芯片.
180
550007
3025
这是我们摄制的薯片袋的高速视频。
09:25
Again it's playing播放.
181
553032
1274
同样,
09:26
There's no chance机会 you'll你会 be able能够
to see anything going on in that video视频
182
554306
3342
您凭肉眼
09:29
just by looking at it,
183
557648
1058
是看不出来什么的,
09:30
but here's这里的 the sound声音 we were able能够
to recover恢复 just by analyzing分析
184
558706
2984
但是我们可以通过分析视频中微小的振动
09:33
the tiny motions运动 in that video视频.
185
561690
2183
恢复出原有的声音。
09:35
(Music音乐: "Mary玛丽 Had a Little Lamb羊肉")
186
563873
2809
(音乐:“Mary Had a Little Lamb”)
09:52
I call it -- Thank you.
187
580985
1486
我叫它 —— 谢谢。
09:54
(Applause掌声)
188
582471
5225
(掌声)
10:01
I call it the visual视觉 microphone麦克风.
189
589878
2345
我叫它视觉麦克风。
10:04
We actually其实 extract提取 audio音频 signals信号
from video视频 signals信号.
190
592223
3390
实际上我们从视频信号中提取了音频信号。
10:07
And just to give you a sense
of the scale规模 of the motions运动 here,
191
595613
3181
为了让您了解运动的大小,
10:10
a pretty漂亮 loud sound声音 will cause原因 that bag
of chips芯片 to move移动 less than a micrometer千分尺.
192
598799
5336
一个特别大的声音可以导致薯片袋一微米的变化。
10:16
That's one thousandth千分之一 of a millimeter毫米.
193
604135
2739
也就是千分之一毫米。
10:18
That's how tiny the motions运动 are
that we are now able能够 to pull out
194
606874
3561
这就是我们能够提取出来的微小运动
10:22
just by observing观察 how light
bounces反弹 off objects对象
195
610435
3243
仅仅是通过摄像机录制的视频
10:25
and gets得到 recorded记录 by our cameras相机.
196
613678
2136
观察光线在物体上的变化。
10:27
We can recover恢复 sounds声音
from other objects对象, like plants植物.
197
615814
3250
我们可以从物体上重构出原声音,如植物。
10:31
(Music音乐: "Mary玛丽 Had a Little Lamb羊肉")
198
619064
6316
(音乐:“Mary Had a Little Lamb”)
10:39
And we can recover恢复 speech言语 as well.
199
627214
1997
我们也可以重构出讲话。
10:41
So here's这里的 a person speaking请讲 in a room房间.
200
629211
2577
这是一个人在房间中讲话。
10:43
Voice语音: Mary玛丽 had a little lamb羊肉
whose谁的 fleece羊毛 was white白色 as snow,
201
631788
4203
(声音:Mary had a little lamb
whose fleece was white as snow,)
10:47
and everywhere到处 that Mary玛丽 went,
that lamb羊肉 was sure to go.
202
635991
4230
and everywhere that Mary went,
that lamb was sure to go.
10:52
Michael迈克尔 Rubinstein鲁宾斯坦: And here's这里的
that speech言语 again recovered恢复
203
640221
2759
迈克尔·鲁宾斯坦:这是一个讲话
10:54
just from this video视频
of that same相同 bag of chips芯片.
204
642980
3274
这段视频同样是由那个薯片袋重构出声音。
10:58
Voice语音: Mary玛丽 had a little lamb羊肉
whose谁的 fleece羊毛 was white白色 as snow,
205
646254
4831
声音:Mary had a little lamb
whose fleece was white as snow,
11:03
and everywhere到处 that Mary玛丽 went,
that lamb羊肉 was sure to go.
206
651085
4859
and everywhere that Mary went,
that lamb was sure to go.
11:07
MR先生: We used "Mary玛丽 Had a Little Lamb羊肉"
207
655944
2346
迈克尔·鲁宾斯坦:我们用儿歌“Mary Had a Little Lamb”
11:10
because those are said to be
the first words
208
658290
2123
因为托马斯 爱迪生在1877年
11:12
that Thomas托马斯 Edison爱迪生 spoke
into his phonograph留声机 in 1877.
209
660413
4161
第一次录制在留声机上的也是这首儿歌。
11:16
It was one of the first sound声音
recording记录 devices设备 in history历史.
210
664574
3228
这是历史上第一个录音设备。
11:19
It basically基本上 directed针对 the sounds声音
onto a diaphragm光圈
211
667802
3327
原理是把声音录在薄膜上
11:23
that vibrated振动 a needle that essentially实质上
engraved the sound声音 on tinfoil锡纸
212
671129
4079
声音通过振动的针头刻录在锡箔上
11:27
that was wrapped包裹 around the cylinder圆筒.
213
675208
2275
锡箔被包裹在圆柱体上。
11:29
Here's这里的 a demonstration示范 of recording记录 and
replaying重播 sound声音 with Edison's爱迪生 phonograph留声机.
214
677483
5943
这是演示爱迪生留声机的录制和回放。
11:35
(Video视频) Voice语音: Testing测试,
testing测试, one two three.
215
683426
3020
(视频)声音:Testing,
testing, one two three.
11:38
Mary玛丽 had a little lamb羊肉
whose谁的 fleece羊毛 was white白色 as snow,
216
686446
3413
Mary had a little lamb
whose fleece was white as snow,
11:41
and everywhere到处 that Mary玛丽 went,
the lamb羊肉 was sure to go.
217
689859
3669
and everywhere that Mary went,
the lamb was sure to go.
11:45
Testing测试, testing测试, one two three.
218
693528
2740
Testing, testing, one two three.
11:48
Mary玛丽 had a little lamb羊肉
whose谁的 fleece羊毛 was white白色 as snow,
219
696268
4156
Mary had a little lamb
whose fleece was white as snow,
11:52
and everywhere到处 that Mary玛丽 went,
the lamb羊肉 was sure to go.
220
700424
5224
and everywhere that Mary went,
the lamb was sure to go.
11:57
MR先生: And now, 137 years年份 later后来,
221
705648
4017
迈克尔·鲁宾斯坦:137年过去了,
12:01
we're able能够 to get sound声音
in pretty漂亮 much similar类似 quality质量
222
709665
4087
我们可以重构出同样质量的声音
12:05
but by just watching观看 objects对象
vibrate颤动 to sound声音 with cameras相机,
223
713752
4079
却是通过观察摄像机拍摄的物体振动实现的,
12:09
and we can even do that when the camera相机
224
717831
1934
我们甚至可以把摄像机
12:11
is 15 feet away from the object目的,
behind背后 soundproof隔音 glass玻璃.
225
719765
4234
放到15英尺外隔音玻璃后边。
12:15
So this is the sound声音 that we were
able能够 to recover恢复 in that case案件.
226
723999
3220
这是我们在这种情况下还原的声音。
12:19
Voice语音: Mary玛丽 had a little lamb羊肉
whose谁的 fleece羊毛 was white白色 as snow,
227
727219
5294
声音:Mary had a little lamb
whose fleece was white as snow,
12:24
and everywhere到处 that Mary玛丽 went,
the lamb羊肉 was sure to go.
228
732513
4759
and everywhere that Mary went,
the lamb was sure to go.
12:29
MR先生: And of course课程, surveillance监控 is
the first application应用 that comes to mind心神.
229
737404
3630
迈克尔·鲁宾斯坦:当然,监控是我们想到的第一个应用。
12:33
(Laughter笑声)
230
741034
2995
(笑声)
12:36
But it might威力 actually其实 be useful有用
for other things as well.
231
744029
4056
但是它应该还有其他用处。
12:40
Maybe in the future未来, we'll be able能够
to use it, for example,
232
748085
2840
可能,未来我们可以用它
12:42
to recover恢复 sound声音 across横过 space空间,
233
750925
2252
太空中还原声音,
12:45
because sound声音 can't travel旅行
in space空间, but light can.
234
753177
3576
因为声音不可以在太空中传播,但光可以。
12:48
We've我们已经 only just begun开始 exploring探索
235
756753
2404
我们仅仅是刚刚开始探索
12:51
other possible可能 uses使用
for this new technology技术.
236
759157
3019
这项新技术的可能用途。
12:54
It lets让我们 us see physical物理 processes流程
that we know are there
237
762176
2832
这让我们熟知的物理过程
12:57
but that we've我们已经 never been able能够
to see with our own拥有 eyes眼睛 until直到 now.
238
765008
3556
变得肉眼可见了。
13:00
This is our team球队.
239
768564
1204
这是我们的团队。
13:01
Everything I showed显示 you today今天
is a result结果 of a collaboration合作
240
769768
2879
今天我展示的一切
13:04
with this great group
of people you see here,
241
772647
2191
都是这群伟大的人的协作成果。
13:06
and I encourage鼓励 you and welcome欢迎 you
to check out our website网站,
242
774838
3167
我鼓励您,欢迎您访问我们的网站,
13:10
try it out yourself你自己,
243
778005
1446
亲身体验,
13:11
and join加入 us in exploring探索
this world世界 of tiny motions运动.
244
779451
2972
加入我们,一同探索微小振动的世界。
13:14
Thank you.
245
782423
1625
谢谢。
13:16
(Applause掌声)
246
784048
1254
(掌声)
Translated by dahong zhang
Reviewed by Donghua Lin

▲Back to top

ABOUT THE SPEAKER
Michael Rubinstein - Research scientist, Google
Computer scientist Michael Rubinstein and his team have developed a "motion microscope" that can show video footage of barely perceivable movements, like breaths and heartbeats.

Why you should listen

Michael Rubinstein zooms in on what we can't see and mangnifies it by thirty or a hundred times. His "motion microscope," developed at MIT with Microsoft and Quanta Research, picks up on subtle motion and color changes in videos and blows them up for the naked eye to see. The result: fun, cool, creepy videos.

Rubinstein is a research scientist at a new Cambridge-based Google lab for computer vision research. He has a PhD in computer science and electrical engineering from MIT.

More profile about the speaker
Michael Rubinstein | Speaker | TED.com