Korea Testing Institute's Self-Driving Car Data Collection and Analysis for Safer Autonomous Vehicles

KTL 자율주행연구센터는 실제 주행 데이터를 수집하고 분석하여 보다 안전한 자율주행 기술 개발에 박차를 가하고 있습니다.

### 레벨4 자율주행차 개발의 현주소: KTL 자율주행 기술 연구센터 탐방
* 🚗 KTL 자율주행 기술 연구센터는 자율주행차 **데이터 확보 및 시뮬레이션 연구**를 통해 자율주행 기술 개발을 지원하는 곳입니다.
* 📸 연구센터는 **4개의 카메라, 3개의 LiDAR, 레이더, 그리고 차량의 속도, 가속도, 위치 정보를 측정하는 프로세스 센서** 등 다양한 센서를 이용하여 데이터를 수집합니다.
* 📊 수집된 **원시 데이터는 서버에서 분석 및 처리되어** 차량, 보행자 등을 인식하고 3D 환경을 재구성합니다.  LiDAR는 고해상도 3D 매핑에 유용하지만 악천후에 취약하며, 레이더는 저해상도이지만 악천후에도 효과적입니다.
* 🚦 데이터 수집 전 **센서 보정 과정**을 거치며, 수집된 데이터는 **다양한 색상으로 구분되어 표시**됩니다 (예: LiDAR 데이터의 색상 구분).  이는 자율주행 시스템의 인식 및 판단 과정을 시각적으로 보여줍니다.
* 🚧 자율주행 테스트 중 **인식 오류 발생 시의 대응**은 중요한 과제이며, 이를 위해 **가상 환경에서 위험 상황을 재현하여 시뮬레이션**합니다.
* ⚠️ 자율주행차 상용화의 가장 큰 걸림돌은 **사고 발생 시 법적 책임 문제**와 **대중의 수용성**입니다. 자율주행차 사고의 대부분은 다른 차량과의 충돌이며, 자율주행차의 사고율은 아직 인간 운전자보다 높습니다.
* 🎯 KTL 자율주행 기술 연구센터의 궁극적인 목표는 **안전 관련 표준을 충족하는 기술 개발 지원**입니다.
* 📈 **방대한 데이터 수집 및 분석**은 자율주행 기술 발전뿐 아니라 **중소기업의 기술 접근성 향상**에도 기여할 것으로 기대됩니다.

If the perception suddenly changes like this... As I said earlier, this is not the kind of equipment that analyzes in real time.

The place I came to is the autonomous driving technology research center of KTL, Korea Testing Institute, which

produces data for evaluating self-driving cars. This is a center. It is not a place that makes self-driving cars,

but a place that secures data that self-driving cars will one day use while driving on the road

and studies it through simulation. You turned it on earlier. Please explain to me about this computer and equipment

. Everything on the pedestrian road. You can think of it as a device that takes pictures of all objects

. The configuration is that the camera you see here has 4 cameras on the front, rear, and sides,

and there is a LiDAR. There are a total of 3 LiDARs. There is a GPS antenna

. LiDAR emits laser (light). Distance is calculated by measuring the time it takes to reflect and return from surrounding objects

. In this process, the 3D environment can be scanned and mapped at high resolution. However, because LiDAR uses light,

it may not be recognized well in bad weather. However, similarly, radar. Radar also emits radio waves and

measures the changes in radio waves that are reflected back from an object, so it can determine not only the distance to the object but also

the speed of the object. Also, since it is a radio wave, it works well even in bad weather. However,

it has a lower resolution than LiDAR, so it can determine the exact speed of the object, but

it can't figure out the shape. There's a sensor inside that can read the speed, acceleration, and location information

of the vehicle. It's called a process sensor, and that sensor also has one inside. All data comes in, and

multiple scenes taken are synchronized to the same time. Before we receive the driving data, we start by calibrating the sensors.

Before collecting the data, we first drive on the road for a bit and calibrate the position and direction of movement of the test car.

As you can see on the screen, everything is displayed in green. Calibration is complete! Now, let's start collecting data right away.

But since this comes in as raw data, it can't be displayed like that. There's a pillar in front.

It can't be displayed like this whether it's a pillar, a cone, or a person.

As I said before, this is not the kind of equipment that analyzes in real time. For analysis, we have a server that receives data

and processes it. If you push it all into the server, it analyzes it and reconstructs it into meaningful data.

If you look at it from that distance, it looks like there's something slightly green, and you

can see it faintly. Even if it works that way, it's actually real. Later,

when we look at it through a viewer or something like that, it recognizes it as a vehicle because there is also data coming from the camera, so we fuse

the two together and analyze it. What do the red and blue colors on the left mean? I told you there are three LIDARs.

I understand that they are colored to distinguish which data comes from which LIDAR. I returned after completing data collection

while driving on the road like this. It also recognizes things that are far away! You're stopping at an intersection, and

there's a car on the other side in the distance. If you look at it, it actually recognizes it. It

automatically classifies cars, pedestrians, etc. Normal cars, like passenger cars, are blue; trucks are red; and

large cars, like buses, are orange. If the perception suddenly changes in the middle like this, what should I do when the autonomous driving test car makes a decision

(during the autonomous driving test)? I think this is what I'm worried about. What I'm worried about is that if something like that happens often,

when my car starts autonomous driving mode and it suddenly moves from one place to another, and the car

recognizes it as such, won't my car be unable to proceed? The scene you see this time

is what it looks like from the camera. While driving on the road, the first straight lane is blocked

by the left-turn lane, and the experimental car moves to the second lane. At this time, if you don't check your rearview properly,

an accident can occur. This situation was observed a while ago. By combining the previously collected lidar data,

a virtual road driving environment is created and used as driving data for self-driving cars. In the virtual world,

what happened in the real environment can be made more dangerous and tested. Automobile manufacturers and national organizations around the world are stepping up.

Why does it take so long for self-driving cars to be commercialized, even though it is necessary to secure and analyze data for self-driving

cars? The biggest obstacle is legal liability in the event of an accident. If an accident occurs in a self-driving car and

it is solely the responsibility of the manufacturer, the manufacturers may abandon self-driving cars. However, they cannot make

it the responsibility of the passengers who do not need to drive.  And the accident rate... there's also the question of acceptability

. What this means is that if you look at the accidents of self-driving cars that are actually being tested,

Most of them occur when a human-driven car collides with the self-driving car. The accident rate

of self-driving test cars is higher than that of human drivers. Even though it is low, ordinary people are afraid. That is why

the accident rate of self-driving cars must be significantly lower than the rate of accidents caused by human drivers

, but it takes longer because people need reassurance. Then, I will ask two questions first.

Among them, the first is about KTL's autonomous driving technology. What is the ultimate goal of the research center?

Related technologies will soon be released in the industry. Once those technologies are released, there will be various requirements from

safety-related standards. Our goal is to support the development of technologies that meet those requirements.

Then, autonomous driving data will be released. I have personally seen how they directly collect data,

turn it into simulations, and process the data. If you collect a lot of data and study it,

what kind of expected effects will this bring to our society as a whole? I think it is related to the goal.

It costs a lot of money to depend on LiDAR, and large companies can afford it,

but it is actually easier for many large and small companies and startups in the overall mobility industry to access this technology. Such companies

can obtain faster and safer data through KTL (Autonomous Driving Technology Research Center). Since KTL provides it without them having to spend time

and resources, it can be used for faster industrial development.  It is our hope that you can contribute even a little bit.

레벨4 자율주행차 되긴 될까? 실제 연구는 어떻게 할까?

현장으로 직접 가는 과학 유튜버 지식인미나니 입니다.
남녀노소가 한 사람의 시선으로 과학을 알아갈 수 있게 현장으로 직접 갑니다.

2020 우수과학도서 저자
 - 책 '요즘 과학 이야기'
 - 책 '내 미래의 직업은?'
 - 책 '요즘 과학' 만화

이력
- KAC한국예술원 특임교수
- 무인탐사연구소 UEL 홍보팀장
- 경기 군포고등학교 명예교사
- 이웃집과학자, 과학동아 객원기자

주요 협업 기관
- IBM, 과학기술정보통신부, 문화체육관광부, 미국국무부, 주한미국대사관, NASA, 한국과학창의재단, 한국항공우주연구원, 한국중부발전, 평창올림픽 등
(채널 소개 자료 제공 가능)

주요 기관행사
- 2022 프랑스 파리 '국제 우주 대회 IAC '
- 2023.01.01 서울시-SBA 제야의 종 인플루언서 
- 2023 IBM-MIT Watson AI Lab 미국 보스턴
- 2023 주한미국대사관 교육 인플루언서 미국 대학 교류
- 2023 누리호 3차 발사 KBS 해설 생중계
- 2023 아제르바이잔 바쿠 '국제우주대회 IAC'

강연/협업 문의 Email : skddl0514@gmail.com 또는 skddl0514@naver.com  (두 곳 다 보내주세요)


지식인미나니

자율주행차 데이터 수집 및 분석 시스템 소개

KTL 자율주행 기술 연구센터 소개: 자율주행차 평가 데이터 생산 및 연구

장비 구성 설명: 카메라(전후좌우 4개), LiDAR(3개), GPS 안테나

LiDAR 및 레이더 작동 원리 및 특징 비교

차량 속도, 가속도, 위치 정보 측정 센서 및 데이터 동기화 과정

센서 보정 및 데이터 수집 과정

원시 데이터 처리 및 서버 분석

서버를 통한 데이터 분석 및 의미있는 데이터 재구성

카메라 및 LiDAR 데이터 융합 및 분석

LiDAR 데이터 색상 구분 및 의미 설명

원거리 물체 인식 및 자동 분류 (자동차, 보행자 등)

자율주행 시험 중 발생 가능한 문제점 및 해결 방안

주행 중 인식 오류 발생 시 대처 방안

인식 오류로 인한 자율주행 차량의 주행 불가능성 문제 제기

카메라 시점에서의 상황 설명 및 사고 위험성

LiDAR 데이터 기반 가상 주행 환경 구축 및 시뮬레이션

자율주행 상용화의 어려움과 KTL의 역할

자율주행차 상용화 지연의 주요 원인: 사고 발생 시 법적 책임 문제

자율주행차 사고 현황 분석 및 안전성 확보 필요성 강조

KTL 자율주행 기술 연구센터의 최종 목표: 안전 관련 표준 충족 기술 개발 지원

데이터 수집, 시뮬레이션, 데이터 처리 과정을 통한 사회적 기대 효과

KTL 연구센터를 통한 중소기업 및 스타트업의 기술 접근성 향상 및 산업 발전 기여