跟讀練習: Turn a Time-of-Flight Sensor into a 3D Scanner - 通過影片學習英語口語

選擇練習模式

每種模式都會開啟全屏練習室。

139 句

如果句子過短或過長，請點選 Edit 進行調整。

Today I want to show you a little setup that I've got here on the breadboard.

These are all parts from Aliexpress so it's a relatively cheap setup.

The ESP cost me about £4, the time of flight sensor cost me about £5 and this IMU was about £15.

So it's the most expensive part of the system and if we wanted we could replace it with a cheaper alternative.

So this whole setup cost about £24, that would be $30 or €28.

Eros.

So the reason why this tiny LiDAR is so special is that it is a multi -zone time -of -flight sensor.

So what that means is that instead of shooting out a single ray

and measuring the distance at which that ray bounces off an obstacle, it shoots out a 8x8 grid.

So those are 64 rays coming out of it that will bounce back once they hit an obstacle.

Now I've connected the breadboard to my computer via a USB cable and the LED is on so we know it's powered.

Here in the viewer now we can see that we've got roughly the same layout for the two breakout boards.

We have the IMU in purple and the time -of -flight sensor in green.

Here on the right we have the status panel showing the live data

and we can see that the IMU is connected

and we're getting measurement data from the time -of -flight at about 15 Hz.

Now you see these blue rays shooting out from the time -of -flight sensor.

There are 64 of them arranged in an 8x8 grid and each represents one zone that the sensor can measure.

Notice how they don't start right at the sensor, there's a small gap.

And they extend outward, but not infinitely.

If I zoom out, you can see that.

These line segments span the sensor's measurable range from 2 cm to 4 m.

Now, this is fundamentally different from how cameras work.

A camera captures light across a continuous image, but the sensor, it can only measure along these specific rays.

The points you see here, they're locked to these line segments.

They cannot appear anywhere else

and you can see as I slide my hand across the sensor you see the points going left

and right and also back and forth

and I can also hold my hand like this

and just slide my finger underneath it

and you can see there's a wave changing the points from the from the palm of my hand

now here's where it gets interesting let me zoom out

and we will see the points here that are projected onto the ceiling of the room where I'm at.

And you see in the settings panel this option here for clipping to measurement.

Let me enable it and let's see what happens.

Do you see how the rays now stop at the measure points instead of extending to the maximum range?

This is a much more accurate representation of what's actually physically happening.

The sensor shoots out infrared laser pulses.

They travel along these paths, bounce off of whatever obstacle they hit, and reflect back.

The sensor measures how long that took, that's the time of flight, and calculates the distance.

So when I hover my hand over the sensor, you can see some rays clip at my hand while others continue to the ceiling behind it.

And that's exactly what's happening in the real world.

Some laser pulses are hitting my hand, others are missing it and hitting the ceiling.

I think this visualization does a much better job of conveying what this sensor does compared to some 2D heatmaps

that you typically see online.

And it's a lot of fun to play with.

Alright, I've zoomed back in because I want to talk a bit about this purple breakout board here.

It's an IMU.

And watch what happens in a visualizer when I rotate the breadboard in the real world.

It also rotates.

The IMU on the breadboard is tracking the orientation in real time

and since both the IMU

and the time -of -flight sensor are stuck to the same breadboard we always know which direction the sensor is pointing.

Let me zoom back out and you can kind of see how if I point around I

can measure different points around my room.

Let's place the breadboard back on the table you may notice

that the points are a bit jittery

and that's just sensor noise from radiation coming from the window or noise from the actual sensor measurement.

Now if I come over here to the side panel

and enable that filter you can see how the noise kind of smooths out

and I can control the strength of the filter with this slider.

What's happening under the hood is an exponential moving average.

At 0 .5 strength, each new position is calculated as 50 % of the new measurement plus 50 % of the previous position.

So it's essentially averaging the current and previous readings.

If I crank up the strength to 0 .9, it gets even smoother.

But now there's a noticeable lag.

So when I move this two -point from the ceiling to the wall, you can see that the points are really smooth,

but there's a noticeable delay.

At 90 % previous and only 10 % new measurement, the visualization can't keep up with the fast movements.

So it's a trade -off between smoothness and responsiveness.

If I move the strength all the way down to 0, that essentially means there is no filter.

It's the same as if this checkbox was left unchecked.

And watch what happens if I move all the way to 1.

That means that the previous measurement gets all the weight and 0 % is assigned to the new measurement.

So even though I'm rotating the breadboard around the room, you see that no new measurements are coming in.

Let me move the slider back to 0 .5, which I found to be a good strength for the filter.

Alright, let's move on to the next section.

The viewer can also fit the plane to the measured points.

When I enable plane fitting with the least squares method and point the sensor at a flat surface like the ceiling,

you can see if it's a nice plane.

But watch what happens when I point the sensor at both the wall and the ceiling at the same time.

The fitted plane ends up diagonally, trying to minimize the error across all points, and this is not very useful.

This is where RANSAC comes in.

RANSAC stands for Random Sample Consensus.

When I switch to this method, watch how it behaves differently.

RANSAC works by randomly picking three of these sampled points, the minimum to define a plane.

Fitting a plane to those three, then checking how many other points lie close to that plane,

and it repeats this about a hundred times and keeps the best fit.

The key difference is, instead of trying to please all the points equally,

Ransack finds the plane that the majority of points agree on and ignores the outliers.

So now when I point again at both the ceiling and the wall, it snaps to whichever surface has more points.

Rather than compromising with a meaningless diagonal.

And so even if I come in with a pen to disturb the sensor measurements,

you can see how it still is able to be robust to that and fit the ceiling nicely.

And if I change to least squares,

you see how much the disturbance from waving this pen on top of the sensor affects the fitted plane.

I'm going to change again to RANSACK now.

If you cover enough points you can see that it still affects the plane fitting algorithm,

but most of the time it behaves much better than least squares.

100

I've changed to least squares again.

101

I'm just hovering the pen back and forth and I'm gonna change to RANSACK again.

102

It's a really big difference.

103

Up until now we've been showing the current snapshot seen by the sensor.

104

In other words, the viewer is only showing the points that the sensor sees right now at any given point in time.

105

When new measurements come in, the old points disappear and that's why it is so interactive.

106

Mapping mode changes that.

107

Instead of replacing points, it accumulates them.

108

So watch what happens when I hover my hand over the sensor with mapping mode enabled.

109

You can see points starting to build up in the shape of my hand's movement,

110

and covering most of the sensors point field of view up close almost like a truncated pyramid.

111

Okay I've zoomed back out and cleared them up.

112

Here's where the magic really happens.

113

You remember we have the IMU tracking orientation right?

114

So watch what happens when I rotate the breadboard.

115

Look at that!

116

The accumulated points are building up a map of the room so I can see the walls, the ceiling, my monitor, my desk,

117

if I go under the desk I can see the floor

118

and if I point at myself you'll even be able to see me

119

and I can even put my left arm up

120

and then my right arm up and

121

because we're no longer just showing the instantaneous points

122

that the sensor is measuring we've just accumulated them we essentially have a map of the room we have a mapping device

123

Now, the reason why this works is because the IMU knows exactly

124

which direction the sensor is pointing at as I rotate it, and each measurement gets placed in the correct position in 3D space,

125

and they all accumulate into a coherent point cloud.

126

This is really cool!

127

With just an ESP32, a time -of -flight sensor, and an IMU, we've essentially built a simple 3D mapping device.

128

Now there's one limitation worth mentioning.

129

The IMU is excellent at measuring rotation, but it cannot accurately track translations, moving the sensors side to side or forward and back.

130

So if I walk around with a breadboard, the map will drift and become inaccurate.

131

For best results, I always try to rotate the sensor in place without translating too much, just like this.

132

Within that constraint, you get surprisingly good results from such a simple and cheap setup.

133

And that's it!

134

This concludes the video.

135

I think this demo really shows how powerful the combination of a multi -zone time -of -flight sensor

136

and an IMU can be.

137

And for around $30 in parts, you can build something that gives you a real intuition for what the sensors are actually measuring

138

and could be useful for a robotics project.

139

Thanks for watching and I'll see you in the next one!

下載應用

流利口語所需的一切

AI 發音評分逐句打分

音標練習掌握每個音

詞彙學習建立你的詞庫

詞彙遊戲邊玩邊學

掃碼

Download on theApp Store Get it onGoogle Play

FOR YOU

本課的詞彙與口語要點

這節口語課以影片“Turn a Time-of-Flight Sensor into a 3D Scanner”為素材。影片中反覆出現的詞有：sensor, Imu, measurement, plane, breadboard。這段影片共有 42 個句子、1785 個單詞可供跟讀。講話部分時長為 11:19。說話人語速自然，每分鐘約 158 個詞，接近日常對話。只有 78% 的單詞屬於英語最常用的 3,000 詞，詞彙難度較高。

影片中的重點詞彙

影片中 15 個值得學習的單詞，附發音和釋義：

單詞	發音	釋義
sensor 名詞	/ˈsɛn.sɚ/	感測器 /感測器
measurement 名詞	/ˈmɛʒ.ə.mənt/	尺寸
ceiling 名詞	/ˈsilɪŋ/	天花板, 天棚
rotate 動詞	/ˈɹoʊ.teɪt/	旋轉, 轉動
essentially 副詞	/ɪˈsɛnʃəli/	本質上 /本質上, 基本上
filter 名詞	/ˈfɪltɚ/	過濾器 /過濾器, 濾過器 /濾過器
smooth 形容詞	/smuːð/	平穩 /平穩, 順利 /順利
viewer 名詞	/ˈvju.ɚ/	觀眾 /觀眾, 看客
obstacle 名詞	/ˈɑbstəkəl/	障礙 /障礙, 妨礙 /妨礙
desk 名詞	/dɛsk/	書桌 /書桌, 寫字檯 /寫字枱 /寫字臺 /寫字臺
connect 動詞	/kəˈnɛkt/	連線 /連線
purple 名詞	/ˈpɝ.pl̩/	紫色
slide 名詞	/ˈslaɪ̯d/	滑梯
segment 名詞	/ˈsɛɡmɛnt/	片斷 /片斷, 段
laser 名詞	/ˈleɪzɚ/	雷射, 雷射

影片中出現的短語動詞

單詞	發音	釋義
hand over 動詞		交出, 移交
zoom out 動詞	/ˌzuːm ˈaʊt/	縮小 /縮小
end up 動詞		到達

值得反覆跟讀的句子

影片中簡短而完整的句子，可以直接用在日常對話裡：

Let me enable it and let's see what happens.
Let's place the breadboard back on the table.
All right, let's move on to the next section.
I've changed to least squares again.

影片中的語法

說話人最常用的結構，並附上影片中的原話：

結構	影片中的用法
條件句 if + 從句，will/would + 動詞 — 條件及其結果	if I point at myself you'll even
現在完成時 have/has + 過去分詞 — 過去發生但與現在仍有關聯的事	I've connected · I've zoomed · I've changed
被動語態 be + 過去分詞 — 強調發生了什麼，而不是誰做的	is connected · are projected · is calculated

需要注意的發音

說話人用了 21 次縮略和弱讀形式，例如 I'm, I've, we've。請按聽到的簡短形式來說。

“th” 音: smooth /smuːð/, underneath /ˌʌndɚˈniθ/, algorithm /ˈælɡəɹɪðm̩/
“sh” 和 “zh” 音: measurement /ˈmɛʒ.ə.mənt/, essentially /ɪˈsɛnʃəli/, orientation /ˌɔː.ɹi.ənˈteɪ.ʃən/, capture /ˈkæp.(t)ʃɚ/, translation /tɹænzˈleɪʃən/
長單詞——注意重音位置: essentially /ɪˈsɛnʃəli/, orientation /ˌɔː.ɹi.ənˈteɪ.ʃən/, representation /ˌɹɛp.ɹə.zɛnˈteɪ.ʃən/, radiation /ˌɹeɪ.diˈeɪ.ʃən/, differently /ˈdɪf.ə.ɹənt.li/

中文母語者容易讀錯的音：

/v/ — 不要讀成 /w/，上齒輕觸下唇: viewer /ˈvju.ɚ/, behave /bɪˈheɪv/, interactive /ˌɪn.tɚˈæk.tɪv/, convey /kənˈveɪ/, whichever /wɪt͡ʃˈɛvɚ/
詞尾子音 — 要讀清楚，後面不要加母音: measurement /ˈmɛʒ.ə.mənt/, rotate /ˈɹoʊ.teɪt/, bounce /baʊns/, desk /dɛsk/, connect /kəˈnɛkt/
/r/ 和 /l/ 的區別: reflect /ɹɪˈflɛkt/, translation /tɹænzˈleɪʃən/, roughly /ˈɹʌf.li/, differently /ˈdɪf.ə.ɹənt.li/, surprisingly /səˈpɹaɪzɪŋli/

如何用這段影片練習

先完整聽一遍影片，不要開口，記下不認識的單詞。
先用 0.75 倍速逐句跟讀，熟練之後再回到正常速度。
錄下自己的聲音並與原聲對比，特別注意 sensor, measurement, ceiling 這類單詞。

什麼是跟讀法？

跟讀法 (Shadowing) 是一種有科學依據的語言學習技巧，最初開發用於專業口譯員的培訓，並由多語言者Alexander Arguelles博士普及。這個方法簡單而強大：您在聽英語母語原聲的同時立即大聲重複——就像是一個延遲1-2秒緊跟說話者的影子。與被動聽力或語法練習不同，跟讀法強迫您的大腦和口腔肌肉同時處理並模模擬實的講話模式。研究表明它能顯著提高發音準確性，語調，節奏，連讀，聽力理解和口語流利度——使其成為雅思口語備考和真實英語交流最有效的方法之一。

影子跟讀法: 閱讀完整分步指南 →

跟讀練習: Turn a Time-of-Flight Sensor into a 3D Scanner - 通過影片學習英語口語

選擇練習模式

下載應用

推薦影片

Making a 360° Controller Smaller than a Penny!

FAST Scan (Focused Assessment with Sonography in Trauma) - STEP by STEP

We Traveled Back in Time. Now Physicists Are Angry.

The Most Overengineered Part of Your iPhone

12 Camera Movements for CINEMATIC FOOTAGE - CREATIVE SHOT IDEAS for BETTER B-ROLL - Video Shot Ideas

Give me 31 minutes, I will make PyTorch click forever

3-D Brain Scans | Pathways 2

Types of Proximity Sensors! #automation #engineering #science #robotics #electronics

Introduction to Optoelectronics and Photonics

PCB 101 Academy - Learn how printed circuit boards are assembled

I Made an EMP Device... It Actually Works

How I Learned Procedural Generation