Application of Digital Tachograph Data for Commercial Vehicle … › system › files › Day3.3.2_Youlim... · 2. Application 1: Prediction for Hazardous Section of GyeongbuExpressway

Application of Digital Tachograph Data for Commercial Vehicle Safetyin Korea

Korea Transportation Safety Authority Cheif Researcher l Dr. Youlim JANG

1. What is Digital Tachograph(DTG)?2. Application 1: Prediction for Hazardous

Section of Gyeongbu Expressway 3. Application 2: COSAS

CONTENTS

2

3

What is Digital Tachograph?

4

What is Digital Tachograph(DTG)?

5

What is Digital Tachograph(DTG)?

6

Application 1: Prediction for Hazardous Section of

Gyeongbu Expressway

7

Data

8

Correlation Analysis

Correlation Analysis between Traffic Accident Rate and Risky Driving Behavior

Speeding

Sudden Acceleration

SuddenDeparture

Sudden Deceleratio

n

Sudden Breaking

Sudden Lane

Changing

Passing

Traffic Accidents

9

Prediction for Risky Driving Behavior

Using Machine Learning(SVM: Support Vector Machine) Model

Digital Tacho Graph Data

Actual Risky Driving Behavior

Potential Risky Driving Behavior

Critical Value

YES NO

YESActual

Risky Driving BehaviorActual

Risky Driving Behavior

NOPotential

Risky Driving BehaviorNormal

Driving Behavior

Act

ual Risky

Drivi

ng B

ehav

ior

Predicted Risky Driving Behavior

10

Prediction of Hazardous Section by Risky Driving Behavior Type

Integrated Data

Bus DTG VDS Volume VDS Avg. Speed Altitude

Risky Driving + Before

3seconds

Risky Driving + Before

5seconds

Analysis Data Pre Processing

Correlation Analysis

Independent Variable : Integrated Data

Dependent Variable : Risky Driving(Y/N)

Continuous Variable : t test Nominal Variable: Chi-Square

Test

Remove unnecessary variables

Standardization Continuous Variable

Generate data for analysis

Risky Driving Data+integrated Data

Risky Driving Behavior Rate1:1

70% Studying 30% Validation

Big Data Analysis

Support VectorMachine

Kernel Trick

Radial Basis

Sigmoid

Linear

Polynomial

Regression Analysis

Ridge Logistic Regression

AnalysisModeling

RBModel

SigmoidModel

LinearModel

PolynomialModel

RidgeModel

Total Integrated Data

Model Validation

Selection Best Model

Calculation Target Rate

Learning by Model

Risky Driving Behavior = Y Data

11

Osan Seoul Osan Seoul

Pridiction Result by Sudden Lane Changing

12

DTGRoad Environment

(Travel Speed etc.)

• Machine Learning Model Performance Comparison

(Machine Learning)• Statistical Analysis• Correlation Analysis• Visualization Scatter Plot• Validation for Pre Processing

Data

(Machine Learning#1)• Input/output variable• Deep learning Model• Time Series Prediction

Model(LSTM) Prediction for

Abnormal Driving Behavior

(머신러닝)• input변수, output변수 결정• 딥러닝 모델 적용• 시계열예측 모델(LSTM) Time

Series Prediction Model(LSTM)

(Machine Learning#2)• Input/output variable• Deep learning Model• Time Series Prediction

Model(LSTM)

HypothesisDrowsy Driving: Abnormal Driving BehaviorAbnormal Driving Behavior: Speeding pattern, RPM, Braking

Pridiction for Hazardous Section with Abnormal Driving Behavior

13

Abnormal Driving Behavior presented

Prediction Results of Driving Behavior using LSTM Deep Learning

Pridiction Result for Hazardous Section by Abnormal Driving Behavior

14

Y (Risk Index)

= 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝐴𝐴𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴𝑆𝑆𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑆𝑆 𝑅𝑅𝐴𝐴𝑅𝑅𝑅𝑅 𝐼𝐼𝑆𝑆𝑆𝑆𝑆𝑆𝐼𝐼 ∗ 0.261

+𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝐷𝐷𝑆𝑆𝐴𝐴𝐴𝐴𝑆𝑆𝐴𝐴𝑆𝑆𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑆𝑆 𝑅𝑅𝐴𝐴𝑅𝑅𝑅𝑅 𝐼𝐼𝑆𝑆𝑆𝑆𝑆𝑆𝐼𝐼 ∗ 0.383

+𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝐿𝐿𝐴𝐴𝑆𝑆𝑆𝑆 𝐶𝐶𝐶𝐴𝐴𝑆𝑆𝐶𝐶𝐴𝐴𝑆𝑆𝐶𝐶 𝑅𝑅𝐴𝐴𝑅𝑅𝑅𝑅 𝐼𝐼𝑆𝑆𝑆𝑆𝑆𝑆𝐼𝐼 ∗ 0.356

Prediction Expressway Speeding

Hazardous Section

Prediction Expressway Sudden

Acceleration Hazardous Section


DeccelerationHazardous Section


Lane Changing Hazardous Section

Finding Expressway Abnormal Driving

Behavior

Prediction Expressway

Abnormal Driving Section

Prediction Expressway

Hazardous Section

Integrating Model

0.261

0.383

0.356

1.0

1)

1)

1)

2)

Mach

ine

Learnin

gM

achin

e Learn

ing

Mach

ine

Learnin

gM

achin

e Learn

ing

Deep

Learn

ing

4)

3)

Pridiction Model for Hazardous Section to Drive

15

Pridiction Results for Hazardous Section to Drive

16

순위 위험 지수 상세 내역 지도

1 0.610

부산기점 403.1km● 위험지수: 0.610● 5년간 사고건수: 24(사망:4,중상:124,경상:22,부상:116)● 과속 위험지수: 0.003● 급가속 위험지수: 0.092● 급감속 위험지수: 0.372● 급진로변경 위험지수: 0.147● 이상행동 위험지수: 0.673

2 0.498


3 0.452

부산기점 403km● 위험지수: 0.452● 5년간 사고건수: 3(사망:1,중상:13,경상:4,부상:9)● 과속 위험지수: 0.005● 급가속 위험지수: 0.091● 급감속 위험지수: 0.261● 급진로변경 위험지수: 0.101● 이상행동 위험지수: 0.494

위험 지수 상세 내역 지도

0.518


0.463


0.414



17

Seoul Toll gateSuwon IC

End of Bus Lane

Suwon ICJamwon IC

Avg

. Risk

Index

Avg

. Risk

Index


18

Total

Company A

Ris

k In

dex


19

Future Application

20

Application 2: COSASConsulting Oriented Safety Assistance System

21

Introduction to COSAS(Consulting Oriented Safety Assistance System)

22

Road Safety Grade Model(1/3)

Road SPF Development

To build SPF, poisson model was selected.

• AADT : annual average traffic volume• length : length of the segment• ddra : sum of 11 hazardous driving behavior from Digital tacho Graph data(2015~2016)• lanes : number of lanes

• EPDO : the predicted traffic accident severity(2015~2016)

23

Road Safety Grade Model (2/3)

Road Safety Grade Model Development

Estimating road safety grade model using SPF-based mean EPDO of reference group.

Road safety grade is divided four grades such as A through D.

24

Road Safety Grade Model (3/3)

Service

25

SSM(surrogate Safety Measure) for Commercial Vehicle Drivers

26

SSM(Surrogate Safety Measure) for Commercial Vehicle Company

27

Service: Commercial Vehicle Company & Driver Safety Grade

28

Future Plan

Q & A

Documents

Application of Digital Tachograph Data for Commercial Vehicle … › system › files › Day3.3.2_Youlim... · 2. Application 1: Prediction for Hazardous Section of GyeongbuExpressway