跟读练习: Kalman Filter - 5 Minutes with Cyrill - 通过视频学习英语口语

选择练习模式

每种模式都会打开全屏练习室。

50 句
如果句子过短或过长,请点击 Edit 进行调整。
1
So what is a Kalman filter?
2
A Kalman filter is a probabilistic state estimation technique and is used to estimate the state of a dynamic system.
3
What is a dynamic system?
4
You can consider a dynamic system a mobile robot moving through the environment
5
and you want to estimate the position orientation of that platform based on motion commands that you send or based on observations,
6
sensor data that you get.
7
Or it can be an autonomous car and then your steering commands, maybe how is your steering angle
8
and how much you are actually pushing your gas pedal
9
and maybe you have a laser range scanner installed on your car in order to observe the environment.
10
This would serve as sensor observations in order to correct your pose.
11
So the Kalman filter basically has two steps, a so -called prediction step and a so -called correction step.
12
The prediction step takes the steering information into account or the motion commands into account in order to estimate,
13
in order to predict where the system will be at the next point in time.
14
So basically it says if I go with one meter per second forward, in one second I will be one meter in front of me.
15
That's what the prediction step does.
16
The correction step takes into account the sensor observation in order to correct for potential mistakes that we have done.
17
So it takes the observation into account and computes the so -called predicted observation.
18
So what should I observe?
19
It would actually be in a state and compares those two observations with each other in order to perform a correction.
20
So for every new command and every new measurement we do a prediction step, correction step.
21
Then the new command, a new observation comes in.
22
So a new prediction step, a new correction step.
23
And so we have this procedure, this recursive procedure, which always updates the previous belief and turns it into the next belief.
24
The Kalman filter makes two important assumptions.
25
The first important assumption is that everything is Gaussian.
26
So we're living in a Gaussian world.
27
That means my belief about the position of the platform is Gaussian and also the uncertainty that I have,
28
for example, in my motion or in my observation.
29
And the second important thing is that all the models are linear
30
that means the model that is involved in the prediction step that estimates where we'll be at the next point in time,
31
and my observation model are both linear models.
32
And these are actually very strong assumptions.
33
If we are, however, in a Gaussian linear world, the Kalman filter is the optimal estimator.
34
So you can't do better in order to estimate the state of a dynamic system.
35
If you're, however, living in a nonlinear world, which is actually the case for most real -world applications, then the Kalman filter can't be used anymore.
36
What you however can do, you can use the so -called extended Kalman filter,
37
which is a variant of the Kalman filter that tries to deal with these non -linearities.
38
And what it does, it basically performs a local linearization.
39
So it performs a Taylor approximation of your non -linear models and turns them into linear models given your current linearization point.
40
So you need to recompute this linearization at every step because you will have a new initial guess,
41
so a new linearization point but then it actually works in the same way as the Kalman filter does,
42
just it adds these local linearizations to that.
43
And the Kalman filter, as well as the extended Kalman filter, is a very popular state estimation technique.
44
It's one of the really basic probabilistic recursive -based filters and is used in a large number of different applications. For example,
45
localizing a vehicle in the environment or tracking the position of the camera as well as local features in the environment.
46
All those state estimation techniques which update the belief of a system giving you data uses those probabilistic state estimation techniques,
47
and the Kalman filter is one of those basic approaches in order to do that.
48
So I hope I gave you a small idea on what the Kalman filter is, what key properties are and what it can be used for,
49
so that you can start diving deeper.
50
Thank you very much for your attention.

下载应用

流利口语所需的一切

AI 发音评分逐句打分
音标练习掌握每个音
词汇学习建立你的词库
词汇游戏边玩边学
FOR YOU

推荐视频

本课的词汇与口语要点

这节口语课以视频“Kalman Filter - 5 Minutes with Cyrill”为素材。 视频中反复出现的词有:filter, Kalman, step, observation, prediction。 这段视频共有 50 个句子、730 个单词可供跟读。 讲话部分时长为 3:53。 说话人语速较快,每分钟约 188 个词,会出现较多连读和弱读。 只有 76% 的单词属于英语最常用的 3,000 词,词汇难度较高。

视频中的重点词汇

视频中 15 个值得学习的单词,附发音和释义:

单词发音释义
filter 名词/ˈfɪltɚ/過濾器 /过滤器, 濾過器 /滤过器
observation 名词/ˌɑbzɚˈveɪʃn̩/觀察 /观察
prediction 名词/pɹɪˈdɪkʃən/預言 /预言, 預估 /预估
estimate 名词/ˈɛs.tɪ.mət/估計 /估计
linear 形容词/ˈlɪn.i.əɹ/线性
correction 名词/kəˈɹɛkʃən/修正, 校正
technique 名词/tɛkˈniːk/技術 /技术, 技巧
belief 名词/bɪˈliːf/相信, 信心
assumption 名词/əˈsʌm(p).ʃ(ə)n/假設 /假设
sensor 名词/ˈsɛn.sɚ/傳感器 /传感器
steer 动词/stɪɹ/掌舵, 驾驶
procedure 名词/pɹəˈsiː.d͡ʒə/做法, 方法
meter 名词/ˈmitəɹ/計量表 /计量表
predict 动词/pɹɪˈdɪkt/預言 /预言
observe 动词/əbˈzɜːv/遵守, 遵循

需要注意的发音

说话人用了 5 次缩略和弱读形式,例如 can't, we'll, we're。请按听到的简短形式来说。

  • “sh” 和 “zh” 音: observation /ˌɑbzɚˈveɪʃn̩/, prediction /pɹɪˈdɪkʃən/, correction /kəˈɹɛkʃən/, assumption /əˈsʌm(p).ʃ(ə)n/, measurement /ˈmɛʒ.ə.mənt/
  • 长单词——注意重音位置: orientation /ˌɔː.ɹi.ənˈteɪ.ʃən/, uncertainty /ʌnˈsɜːtənti/, autonomous /ɔˈtɑnəməs/, estimation /ˌɛs.təˈmeɪ.ʃən/, probabilistic /ˌpɹɑ.bə.bɪˈlɪs.tɪk/

中文母语者容易读错的音:

  • /v/ — 不要读成 /w/,上齿轻触下唇: observation /ˌɑbzɚˈveɪʃn̩/, observe /əbˈzɜːv/, dive /daɪv/, variant /ˈvæɹi.ənt/, recursive /ɹɪˈkɜː(ɹ)sɪv/
  • 词尾辅音 — 要读清楚,后面不要加元音: estimate /ˈɛs.tɪ.mət/, technique /tɛkˈniːk/, belief /bɪˈliːf/, dynamic /daɪˈnæm.ɪk/, extend /ɛkˈstɛnd/
  • /r/ 和 /l/ 的区别: linear /ˈlɪn.i.əɹ/, probabilistic /ˌpɹɑ.bə.bɪˈlɪs.tɪk/, linearity /ˌlɪniˈæɹəti/

如何用这段视频练习

  1. 先完整听一遍视频,不要开口,记下不认识的单词。
  2. 先用 0.75 倍速逐句跟读,熟练之后再回到正常速度。
  3. 录下自己的声音并与原声对比,特别注意 filter, observation, prediction 这类单词。

什么是跟读法?

跟读法 (Shadowing) 是一种有科学依据的语言学习技巧,最初开发用于专业口译员的培训,并由多语言者Alexander Arguelles博士普及。这个方法简单而强大:您在听英语母语原声的同时立即大声重复——就像是一个延迟1-2秒紧跟说话者的影子。与被动听力或语法练习不同,跟读法强迫您的大脑和口腔肌肉同时处理并模仿真实的讲话模式。研究表明它能显着提高发音准确性,语调,节奏,连读,听力理解和口语流利度——使其成为雅思口语备考和真实英语交流最有效的方法之一。

影子跟读法: 阅读完整分步指南 →