Lesani, Miranda-Moreno, Fu, Romancyshyn 1 DEVELOPMENT AND TESTING OF AN ULTRASONIC-BASED 1 PEDESTRIAN COUNTING SYSTEM 2 3 4 Asad Lesani, Ph.D. student (Corresponding author) 5 Department of Civil Engineering and Applied Mechanics, McGill University 6 Room 492, Macdonald Engineering Building, 817 Sherbrooke Street West 7 Montreal, Quebec H3A 2K6 8 Tel: +1 (514) 473-4292 9 Email: [email protected]10 11 Luis F. Miranda-Moreno, Ph.D., Associate Professor 12 Department of Civil Engineering and Applied Mechanics, McGill University 13 Room 268, Macdonald Engineering Building, 817 Sherbrooke Street West 14 Montreal, Quebec H3A 2K6 15 Tel: +1 (514) 398-6589 16 Fax: +1 (514) 398-7361 17 Email: [email protected]18 19 Ting Fu, Ph.D. student 20 Department of Civil Engineering and Applied Mechanics, McGill University 21 Montreal, Quebec H3A 2K6 22 Tel: +1 (514) 398-6589 23 Email: [email protected]24 25 Taras Romancyshyn, Undergraduate Research Assistant 26 Department of Civil Engineering and Applied Mechanics, McGill University 27 Room 391, Macdonald Engineering Building, 817 Sherbrooke Street West 28 Montreal, Quebec H3A 2K6 29 Email: [email protected]30 31 32 Word count 33 Text 4467 Tables ( X 250) 1 Figures ( X 250) 11 Total 7467 34 Paper prepared for presentation at the 94 nd Annual Meeting of the Transportation Research Board, 35 January 2015 36 37
17
Embed
DEVELOPMENT AND TESTING OF AN ULTRASONIC-BASED PEDESTRIAN COUNTING SYSTEM
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Lesani, Miranda-Moreno, Fu, Romancyshyn
1
DEVELOPMENT AND TESTING OF AN ULTRASONIC-BASED 1
PEDESTRIAN COUNTING SYSTEM 2
3 4
Asad Lesani, Ph.D. student (Corresponding author) 5 Department of Civil Engineering and Applied Mechanics, McGill University 6 Room 492, Macdonald Engineering Building, 817 Sherbrooke Street West 7 Montreal, Quebec H3A 2K6 8 Tel: +1 (514) 473-4292 9 Email: [email protected] 10
11
Luis F. Miranda-Moreno, Ph.D., Associate Professor 12 Department of Civil Engineering and Applied Mechanics, McGill University 13 Room 268, Macdonald Engineering Building, 817 Sherbrooke Street West 14 Montreal, Quebec H3A 2K6 15 Tel: +1 (514) 398-6589 16 Fax: +1 (514) 398-7361 17 Email: [email protected] 18
19
Ting Fu, Ph.D. student 20 Department of Civil Engineering and Applied Mechanics, McGill University 21 Montreal, Quebec H3A 2K6 22 Tel: +1 (514) 398-6589 23 Email: [email protected] 24
25
Taras Romancyshyn, Undergraduate Research Assistant 26 Department of Civil Engineering and Applied Mechanics, McGill University 27
Room 391, Macdonald Engineering Building, 817 Sherbrooke Street West 28 Montreal, Quebec H3A 2K6 29
4.3 Results 3 The general results of the three sites are presented in Table 1. From this table, one can 4
observe that the errors percentage varied between 0.9 and 24.7 for the ultrasonic sensor and 5
between 0.8 and 42.1 for the infrared sensor over the 3 test sites. In addition, the average AADP 6
for the ultrasonic and infrared sensors ranged from 6.4 to 12.3 and 4.6 to 19.4, respectively. 7
These results clearly show that the Ultrasonic sensor performed better. 8
9
10
11
12
13
14
15
Lesani, Miranda-Moreno, Fu, Romancyshyn
12
1
2
Table 1. Summary statistics of tests per site 3
Measure Site 1 Site 2 Site 3
Error (%)
Infrared Min - Min 0.9 Min 0.8
Max - Max 42.1 Max 9.9
Ultrasonic Min 1.7 Min 3.7 Min 0.9
Max 24.7 Max 18.6 Max 27.2
ADP (%)
Infrared - -19.4 -0.5
Ultrasonic -11.3 -9.9 2.8
AADP (%)
Infrared - 19.4 4.6
Ultrasonic 12.3 9.9 6.4
4
5
Figures 9a and 9b show the automated and manual counting values (46 intervals of 15-6
min counts) for the first site. Figure 9.a shows the manual versus (vs) the automatic ultrasonic 7
counts from which an R2 of 0.88 was obtained. Figure 9.b shows the error measures. As 8
expected, an under-counting problem was observed, which increased slightly as volumes 9
increased. The system over-counted pedestrians in only one case. 10
11
Figure 10.a and 10.b show the results for the second site (Sherbrooke). This site had 12
slightly greater volumes than Milton. Moreover, automated pedestrian counts from both Eco-13
Counter infrared and ultrasonic sensors are presented and compared with respect to manual 14
counts obtained from video data. These results clearly show again the better performance of the 15
ultrasonic system compared than the infrared system. The ultrasonic system presents much lower 16
errors and a higher R2 measure. Also, the results of site 2 show error values of more than 25% 17
several times. This highlights the limitation of the infrared based sensor in sidewalks that are 18
located in open areas. The infrared sensor needs an obstruction, such as wall, or well-defined 19
detection area to have accurate counting. On the other hand, the error values of the counting with 20
the ultrasonic system are less than 15% most of the time. Also, one of the advantages of the 21
ultrasonic sensor is that the measurement range of the ultrasonic sensor can be limited being able 22
to be used in an open area without any obstruction. 23
24
Figure 11.a and 11.b show the result of the third site on University Street. Figure 11.a 25
shows the manual vs the automatic ultrasonic and automatic infrared sensors and Figure 11.b 26
shows the error values. From these figures, it can be seen that most of the time the error values of 27
counting with both systems lie between 0 to 12%. This shows a relatively high level of accuracy 28
in both systems, with very high R2 in both cases. However, there are some over-counting issues 29
because of the high pedestrian volumes as well as the stop-and-going pattern of the pedestrian 30
Lesani, Miranda-Moreno, Fu, Romancyshyn
13
flow. That is, the over-counting error is because pedestrians stop for certain amounts of time in 1
front of the sensors. The over-counting error seems to be more severe in the ultrasonic sensor 2
under this stop-and-going condition. 3
4
a) Site 1. Manual vs Ultrasonic counts
b) Site 1 – Ultrasonic error
5
Figure 9. Test outcomes for site 1 6
-30
-25
-20
-15
-10
-5
0
5
10
15
20
25
50 70 90 110 130 150 170
Counts
(%)
Volume (15-min periods)
Lesani, Miranda-Moreno, Fu, Romancyshyn
14
1
2
3 a) Site 2. Ultrasonic vs Infrared counts 4
5 b) Site 2 – Ultrasonic vs Infrared errors 6
Figure 10. Test outcomes for site 2 7
8
Lesani, Miranda-Moreno, Fu, Romancyshyn
15
1
2 a) Site 3. Ultrasonic vs. Infrared counts 3
4 a) Site 2 – Ultrasonic vs Infrared errors 5
Figure 11. Test outcomes for site 3 6
Lesani, Miranda-Moreno, Fu, Romancyshyn
16
1
CONCLUSIONS AND FUTURE WORK 2 3
This paper introduces an original automated pedestrian counting system based on 4
ultrasonic technology. The system components, hardware and software, are described in this 5
document along with the testing protocol. Three different sites were selected to test the accuracy 6
of the counting system and its functionality in different challenging situations including high 7
volumes, large counting areas without a wall present and stop-and-going pedestrian flow 8
conditions. Video data was collected in order to obtain manual counts (defined as the “ground 9
truth”). Sites and counting periods were selected in order to observe a large variability in the 10
magnitude of counts. One of the sites had a wide sidewalk without a wall, in which traditional 11
counting technologies do not work properly. 12
13
The initial findings clearly show the potential of this counting system with a very good 14
performance on sidewalks with low to high volumes (from 200 to 800 pedestrians per hour). The 15
under-counting error, due mainly to occlusion, varies between 0 to -20% in 98% of the cases. 16
Based on the results, the error increases slightly with volumes, with a power or linear function, 17
which can be used to generate correction functions. These functions can be used to improve the 18
counting estimates (reduce error) in congested sidewalks. 19
20
The proposed system could help handling the limitations of infrared technology, such as 21
the requirement of an obstruction, such as a wall, or well-defined detection area. In open areas or 22
wide sidewalks without walls, the definition of the counting area is a challenge for infrared 23
sensors. In this condition, the accuracy of the infrared based sensor is deteriorated. In the 24
proposed ultrasonic system, there exists the option of changing the coverage area, so it can be 25
used at different locations including open spaces. Additionally, the ultrasonic technology is not 26
sensitive to temperature as opposed to an infrared based sensor which can be significantly 27
affected on hot-sunny days. 28
29
One of the main limitations of the ultrasonic sensor with respect to infrared sensor is the 30
higher energy consumption. Batteries in infrared sensors can work for several years, while in our 31
system the battery only works for a few days. In addition, there a number of interesting 32
improvements and developments that must be explored to further enhance our product. These 33
include: 34
• Modify the detection methodology to count the number of cyclists in bicycle facilities. 35
• Reduce the power consumption of the system to increase the lite time of the system 36
working on battery. 37
• Modify the system to count pedestrians based on the direction of travel (using two 38
sensors in parallel). 39
• Implementing a system with two ultrasonic sensors to increase the accuracy of distance 40
measurements and decrease measurement noises. 41
• Implement data fusion theory in order to create a system that works alongside other 42
sensors such as a combination of a Pyro-electric Infrared sensor and an ultrasonic sensor. 43
• Test the system under different weather conditions. 44
45
46
Lesani, Miranda-Moreno, Fu, Romancyshyn
17
1
2
REFERENCES 3 1. Greene-Roesel, Diogenes, Ragland and Lindau, “Effectiveness of a Commercially Available 4
Automated Pedestrian Counting Device in Urban Environments: Comparison with Manual 5 Counts”, Safe Transportation Research & Education Center, 2008 6
2. Ismail, Karim, Tarek Sayed, and Nicolas Saunier. "Automated collection of pedestrian data using 7 computer vision techniques." Transportation Research Board Annual Meeting Compendium of 8 Papers, Washington, DC. 2009. 9
3. Markowitz, Frank, Jeannette Montufar, and Maury Steindel. "Automated Pedestrian Detection: 10 State of the Art." ITE 2009 Annual Meeting and Exhibit. 2009. 11
4. Dharmaraju Raghuram, David A. Noyce, and Joshua D. Lehman. An evaluation of technologies 12 for automated detection and classification of pedestrians and bicycles. Diss. University of 13 Massachusetts at Amherst, 2002. 14
5. Leykin, Alex, and Riad Hammoud. "Robust multi-pedestrian tracking in thermal-visible 15 surveillance videos." Computer Vision and Pattern Recognition Workshop, 2006. CVPRW'06. 16 Conference on. IEEE, 2006 17
6. Greene-Roesel, Ryan, et al. "Effectiveness of a Commercially Available Automated Pedestrian 18 Counting Device in Urban Environments: Comparison with Manual Counts." (2008). 19
7. Ozbay, K., H. Yang, and B. Bartin, “Calibration of An Infrared-based Automatic Counting 20 System for Pedestrian Traffic Flow Data Collection”, Transportation Research Board 89th Annual 21 Meeting, 2010, 10-3574. 22
8. Leibe, Bastian, Edgar Seemann, and Bernt Schiele. "Pedestrian detection in crowded scenes." 23 Computer Vision and Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society 24 Conference on. Vol. 1. IEEE, 2005. 25
9. Krammer, Peter, and Herbert Schweinzer. "Localization of object edges in arbitrary spatial 26 positions based on ultrasonic data." Sensors Journal, IEEE 6.1 (2006): 203-210. 27