Searching for severity dimension among traffic events Oksana - - PDF document

Oksana Yastremska Lund University, Sweden

Searching for severity dimension among traffic events

32nd ICTCT Conference in Warsaw, Poland October 25, 2019

SURROGATE MEASURES OF SAFETY

"Safety pyramid", (adopted from Hyden, 1987) 25/10/19 | 2

SOME SURROGATE SAFETY INDICATORS:

25/10/19 | 3

RESEARCH QUESTION:

Which objective parameters provide ‘acceptable’ safety gradation of traffic events?

25/10/19 | 4

Nearness

25/10/19 | 5 Conceptual illustration of severity, Laureshyn et al., (2017)

Indicators (features) Human judgements MODEL

25/10/19 | 6

METHOD:

HUMANS’ JUDGEMENTS:

25/10/19 | 7 25/10/19 | 8

‘Car – car’ vs. ‘Car – car’ ‘Car – cyclist’ vs. ‘Car – cyclist’ ‘Car – car’ vs. ‘Car – cyclist’

COMPARISON GROUPS:

Indicators (features) Human judgements Model

25/10/19 | 9

Initial conditions Culmination Yes No

25/10/19 | 10

Evasive action

Initial conditions Culmination Intensity of evasive action

• T2;
• distanc

e;

• DeltaV0

;

• E
• Decelerati
• n
• T2min;
• PET;
• distanc

e;

• DeltaV0

;

• E

25/10/19 | 11

FEATURES:

25/10/19 | 12

Indicators (features) Human judgements Model LOGISTIC REGRESSION

25/10/19 | 13

More severe Binary classifier A B A 1 C D D E F E 1 G H H … … … … … … … … … … … … Pair ... A B C … T2 ... 1,15 1,63 0,89 ... DeltaV0 (T2) ... 4,9 5,3 5,4 ... Emax (T2) ... 208,809 250,9 250,895 ... Distance (T2) ... 8,66175 9,9532 7,9734 ... T2min ... 3,34378 0,4643 2,92312 ... DelvaV0 (T2min) ... 2,6 6,9 1,5 ... Emax(T2min) ... 60,5422 422,87 19,7689 ... Distance (T2min) ... 6,93876 4,0455 4,63004 ... PET ... 1,15783 0,3582 2,54086 ...

• Av. Deceleration

... 5,6475 0,7298 4,23107 ...

XA=

T

A 1,15 4,9 208,81 8,6618 3,3438 2,6 60,542 6,9388 1,1578 5,6475

XB=

B 1,63 5,3 250,9 9,9532 0,4643 6,9 422,87 4,0455 0,3582 0,7298

T HUMAN JUDGEMENTS FEATURES

lin(x)=wx+b

S(XA) > S(XB) if w XA >w XB

25/10/19 | 14

ANALYSIS OF EACH FEATURE SEPARATELY:

25/10/19 | 15

CORRELATIONS OF ALL INDICATORS TOGETHER:

| 16

ANALYSIS OF EACH FEATURE GROUP SEPARATELY:

Moment of initial conditions Intensity of evasive action Moment of ‘culmination’

25/10/19 | 17

• No clear winner or loser among single indicators
• Combinations perform better than single indicators
• Still 20 % are not explained – why?

First conclusions:

• More indicators
• More training data
• Inter-observer reliability

Oksana Yastremska Transport & Roads Department of Technology & Society Faculty of Engineering, LTH Lund University

• ksana.yastremska@tft.lth.se

25/10/19 | 18