automated road safety analysis and data …...results ismail, sayed and saunier (2010)...
Post on 09-Jul-2020
0 Views
Preview:
TRANSCRIPT
Automated Road Safety Analysis
and Data Collection Using
Video Sensors
Tarek Sayed
University of British Columbia
Acknowledgement
Nicolas Saunier (École Polytechnique de Montréal)
Karim Ismail (Carleton University)
Mohamed Zaki (UBC)
Jarvis Autey (UBC)
Greg Mori (SFU)
Engineering Approaches to Road
Safety Management
Reactive approach Making improvements to existing unsafe road locations based on accident history
Proactive approach Prevent unsafe road conditions from occurring by including road safety as a priority at the planning/ design and early operation stages
The earlier that road safety is considered, the more cost-effectively it can be accommodated
Motivation
Traditional road safety analysis is a reactive approach, based on historical collision data
There are well-recognized availability and quality problems associated with collision data Less complete understanding of the complex interaction of collision factors and how safety measures work A more proactive approach is needed which provides a better understanding of collision occurrence
Traffic Conflicts (near-misses)
Shortcomings Cost of data collection Issues related to the reliability and accuracy of human observers
Source: Hyden, 1987
A Modular System for Vision-based
Automated Road Safety Analysis
Motion PatternsVolume, Origin-
Destination CountsDriver Behavior...
Trajectory Database Interaction Database
Traffic Conflict DetectionExposure MeasuresInteracting Behavior...
Image Sequence
Interpretation Modules
(Saunier and Sayed, 2006)
Video Analysis
Real-world Coordinates Recovery
1 2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
1
2
3
4
5 6
7 8 9 1
0 11
12
13
14
15
16
17
18
19 2
0
21
22
(Ismail, Sayed and Saunier, 2009)
Recovery of Real-world
Coordinates
o Calgary Glenmore Trail & 5 Street
o Edmonton
Recovery of Real-world
Kuwait City
Kuwait City
Video Analysis
Example of Motion Patterns (Calgary)
Calgary Gilmore and No. 5
Video Analysis
Road User Classification
Oakland, CA Chinatown 2004
Oakland, CA Chinatown
(Ismail, Sayed and Saunier, 2009)
Video Analysis
Road User Classification
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
% false positive rate (veh as ped)
% t
rue p
ositiv
e r
ate
(ped a
s p
ed)
Receiver operating characteristic curve for three classification schemes
Max speed classifier
Smoothed max speed classifier
Prototype classifier
Ismail, Sayed and Saunier (2010)
Video Analysis
Validation
Oakland, CA Chinatown 2004
Vancouver Fire Works
Vancouver Fire Works
(with Greg Mori SFU)
- - Canadian Conference on Computer and Robot Vision (CRV 2010)
Objective Conflict Indicator
Vehicle-vehicle Interactions
Ux is current track Hxy is extrapolation hypothesis CP
x is collision point
U1
U2
H11
H12
H21
H22
CP2
CP1
(Saunier and Sayed, 2008)
Results (Old Training Video)
Results
Conflict Indicators calculated for traffic event 1.
DST: Deceleration t(Ismail, Sayed and Saunier, 2010)o Safety Time, GT: Gap \time, TTC: Time to
Collision, PET: Post Encroachment Time
Automated Before-and-After Safety
Projects
1. Before/After Evaluation of Pedestrian Scramble (California)
2. Before/After Analysis for the Treatment at Yellowhead / Victoria Trail
Before/After Evaluation of Pedestrian
Scramble (Ismail, Sayed and Saunier, 2010)
Motion Patterns Pre-Scramble Post-Scramble
Results
Ismail, Sayed and Saunier (2010)
Before-and-After Conflict Indicators
0 2 4 6 8 10 12 14 16 18 200
100
200
300
400
500
600Histogram of Before-and-After TTC
TTCmin
in seconds
Fre
quency o
f tr
aff
ic e
vents
-10 -8 -6 -4 -2 0 2 4 6 8 100
2000
4000
6000
8000
10000
12000
DSTmax
in seconds
Fre
quency o
f tr
aff
ic e
vents
Histogram of Before-and-After DST
-20 -15 -10 -5 0 5 10 15 200
500
1000
1500
2000
2500
3000
PET in seconds
Fre
quency o
f tr
aff
ic e
vents
Histogram of Before-and-After PET
0 2 4 6 8 10 12 14 16 18 200
1000
2000
3000
4000
5000
|PET| in seconds
Fre
quency o
f tr
aff
ic e
vents
Histogram of Before-and-After |PET|
-5 0 5 10 15 200
1000
2000
3000
4000
GTmin
in seconds
Fre
quency o
f tr
aff
ic e
vents
Histogram of Before-and-After GT
0 2 4 6 8 10 12 14 16 18 200
1000
2000
3000
4000
|GTmin
|in seconds
Fre
quency o
f tr
aff
ic e
vents
Histogram of Before-and-After |GT|
TTC Before
TTC After
DST Before
DST After
PET Before
PET After
|PET| Before
|PET| After
GT Before
GT After
|GT| Before
|GT| After
Ismail, Sayed and Saunier (2010)
B/A Studies (Scramble Phase)
Before After
Ismail, Sayed and Saunier (2010)
EDMONTON B/A YELLOWHEAD
/ VICTORIA TRAIL RAMP
Motion Patterns
Motion Patterns
Camera Calibration (Before)
Conflict Examples (Before Data)
Conflict Examples (Before Data)
After Analysis
Conflict Examples (After Data)
Results
CONFLCIT TYPE
BEFORE
TREATMENT AFTER TREATMENT
B/A RATIO
Merging
1.3151e-004 4.5799e-006 28.71
1.1399e-004 4.1768e-006 27.29
9.1363e-005 3.4317e-006 26.62
Rear-end
2.8288e-005 2.6896e-005 1.05
2.8716e-005 2.3093e-005 1.24
2.4232e-005 1.6991e-005 1.43
All Types
9.1685e-005 7.6311e-006 12.01
8.1091e-005 6.7632e-006 11.99
6.5463e-005 5.2856e-006 12.39
Penticton, BC - Before/After
Analysis of Right-turn Treatment
Evaluating the Reliability of Red-
Light Cameras
Evaluating the Reliability of Speed
Cameras
Automated Safety Analysis -
Conclusion A new approach to road safety analysis
Proactive, generic and low cost approach
Provides better understanding of driver behavior especially collision avoidance mechanisms
Diagnostic approach
Overcomes the problems with the traffic conflict technique (high cost and reliability of observers)
It is time to take safety analysis in a new direction
Ongoing Studies /Future Work
Automatic detection of traffic violation events (TRB, 2011)
Several B/A studies using conflicts
Aggregation of conflict indicators (TRB, 2011)
Safety diagnosis of collision prone locations (roundabout)
Conflict/Collision models
top related