A Tracking-based Traffic
Performance measurement System for
Roundabouts/Intersections
PI: Hua Tang
Graduate students: Hai Dinh
Electrical and Computer Engineering Department
University of Minnesota Duluth
Proposed FY11 project
• The goal is to develop a tracking-based traffic
performance measurement system for roundabouts
and intersections
• Performance measurements are in great need for
roundabouts and intersections, which are more
challenging than highways (more complex
measurement specifications and vehicle behavior)
• Many traditional techniques (using sensors) can only
obtain partial performance measurements
[1] T. Kwon, “Portable Cellular Wireless Mesh Sensor Network for Vehicle Tracking in an Intersection”, Minnesota
Department of Transportation, Final report, CTS 08-29.
[2] A. Abdel-Rahim, B. Johnson, “An Intersection Traffic Data Collection Device Utilizing Logging Capabilities of Traffic
Controllers and Current Traffic Sensors”, National Institute for Advanced Transportation Technology, Final report.
Basic project information
• Traffic performance measurements can be processed
offline
• Traffic performance measurements for roundabouts and
intersections include:
–
–
–
–
–
–
vehicle volume
vehicle speed (including acceleration/de-acceleration behavior)
origin-destination pairs
waiting time
gap size (by turning or entering vehicles)
lane use, etc
• To apply video/image processing techniques based on
probabilistic tracking algorithms to individually track
each vehicle from its entrance to exit
Step 1: camera calibration
Actual camera height:
51 feet
Calibrated camera height: 52.4 feet
The height of the camera is 51 feet, which is close to our result (52.36 feet).
Step 2: vehicle tracking for trajectory
The height of the camera is 51 feet, which is close to our result (52.36 feet).
Step 3: data mining for measurements
Lane 1
Number of Vehicle
Start Time
1:00:00 PM
1:00:30 PM
1:01:00 PM
1:01:30 PM
1:02:00 PM
1:02:30 PM
1:03:00 PM
1:03:30 PM
1:04:00 PM
Intersection
Frame
92
133
136
229
248
257
444
1295
1416
Enter
7
3
2
0
4
4
7
4
16
Exit
4
3
4
1
3
3
7
6
15
Time
minute second
0
13
0
19
0
19
0
33
0
35
0
37
1
3
3
5
3
22
Average
Travel
Remain
Time
3
15.83333
3
12.66667
1
17.03333
0
13.86667
1
17.15556
2
11.77778
2
16.48889
0
12.73333
1
10.13333
gap size
Lane
Upper or
(s)
Relation Lower
2.285714
21
1
3.714286
31
2
3.285714
31
1
6.857143
21
2
4.142857
21
2
2.857143
21
1
5.285714
21
2
4.285714
21
2
4.714286
21
2
Start
Frame
End
Frame
Travel
Time
93
165
180
193
366
382
397
475
682
957
971
212
258
274
291
526
549
479
577
786
1100
1100
15.86667
12.4
12.53333
13.06667
21.33333
22.26667
10.93333
13.6
13.86667
19.06667
17.2
Lane 1
Starting position
x
y
143
136
136
142
134
137
145
146
142
140
143
620
627
628
624
631
629
613
590
600
622
620
Ending Position
x
y
208
212
211
216
212
210
206
211
212
216
216
9
9
10
10
11
9
11
9
9
8
7
The height of the camera is 51 feet, which is close to our result (52.36 feet).
20
15
10
5
Number of vehicle entering ramp 1
Number of vehicle entering ramp 1
Results I
0
2:30 2:40 2:50 3:00 3:10 3:20 3:30 3:40 3:50 4:00
Time (PM)
Number of vehicle exiting ramp 1
Number of vehicle exiting ramp1
5
10
5
4:20
4:40
5:00
5:20
Time (PM)
4:20
4:40
5:00
5:20
Time (PM)
5:40
6:00
20
20
10
15
0
4:00
0
2:30 2:40 2:50 3:00 3:10 3:20 3:30 3:40 3:50 4:00
Time (PM)
15
20
15
10
5
0
4:00
5:40
6:00
The height of the camera is 51 feet, which is close to our result (52.36 feet).
Results II
20
35
15
Number of occurences
Waiting time (second)
30
10
5
25
20
15
10
5
0
0
20
20
15
15
10
5
0
4:00
4:20
4:40
5:00
5:20
Time (PM)
5:40
6:00
5
10
15
10
15
Gap size (s)
Number of occurences
Waiting time (second)
0
2:30 2:40 2:50 3:00 3:10 3:20 3:30 3:40 3:50 4:00
Time (PM)
10
5
0
0
5
Gap size (s)
The height of the camera is 51 feet, which is close to our result (52.36 feet).
Computing gap size
(1) For each vehicle, detect when it enters
(2) Record the position of each other vehicle
(3) Compute the distance or travel time
Summary and plan
• We have tested more than 50 videos so far.
• The processing time to video length ratio is about 2.
• The vehicle count accuracy and gap size accuracy are
over 95% and waiting time accuracy are over 92%.
• The system is subject to more test (origin-destination
pairs) and needs to be improved for better accuracy.
The height of the camera is 51 feet, which is close to our result (52.36 feet).
A commercial system
The height of the camera is 51 feet, which is close to our result (52.36 feet).
Another commercial system
Acknowledgement
•
•
•
•
NATSRL
Dr. Eil Kwon (NATSRL)
District I office
MN-DoT