How to count people entering a shop?

PeterH_DK

Member
Join Date
Jan 2006
Location
Copenhagen
Posts
17
I have set up some polling devices in retail shops using a PLC ( Trilogy T100MD888) and a GSM modem to transmit data.

It works fine, but the customers would also like to count how many people enter the shops. At first I installed a Through-beam photoelectric sensor from SICKin front od the door, and divided the counted pulses by two assuming that everybody who enters also exits. This has worked for three months but required frequent attention because either the light beam or the reflector were bumped into by customers and thus got out of line and failed to count..

I therefore tried with a small PIR sensor from Parallax. The sensor is mounted in a 10cm long tube with a rectangular hole of about 2 * 8 cm and hangs from the ceiling over the entry door. It therefore only detects movement in the doorway.

In order to detect the reliability I mounted both sensors in one shop and could see that at the end of each day the registered amount of visitors was almost identical, the difference in counting was less than 3%. At first I was very happy: 3% is fine. We are not looking for a high accuracy like we could get with a Video camera – rather for a very cheap solution, that can indicate the customer flow per hour in the shops.



But when I analyze the actual counting I see a difference between the two counting methods. It levels out during the day - sometime one sensor registers more, sometimes the other.
I have reduced the counting resolution in order to save space. I have done this by counting the sensor pulses and then writing them to memory with a timestamp every time somebody votes on the polling device. This means I only know how many customers entered since the last vote. The resolution is acceptable because I collect 2-300 votes per day corresponding to approx 20% of the customers entering the shop.


But seeing the difference in measurement from the two devices it is clear that I do not count everybody who enters the shop and there are several reasons for this.

Firstly if two people walk in hand-in-hand, or two people enter and exit at the same time, that will only generate one pulse. Or if somebody stops in the doorway counting is blocked for as long as he or she is there.

So I am definitely registering less people than have actually have entered.



The question how much less? I have contemplated standing there for a couple of hours and manually registering how many entered the shop. I would then be able to estimate how many I missed



But before I stand there being totally bored for several hours I would like to hear if any of you could recommend a different and cost effective way of measuring how many people enter a shop on an hourly basis. Cost and ease of installation are both important issues. So cameras are out. But is there a better way?

/Peter
 
How about some foot traffic counting method? Counting feet instead of bodies may help. Kind of like a traffic flow test where they put the sensor across the street and it counts how many wheels pass over it.
 
Tharon said:
How about some foot traffic counting method? Counting feet instead of bodies may help. Kind of like a traffic flow test where they put the sensor across the street and it counts how many wheels pass over it.

Interesting yes - but I still have the problem og several being on the "activator" at the same time, unless i could have a mat with several sensors - haven't heard of any though.

/Peter
 
PeterH_DK

You can still use a through beam. And condition your logic to "look" for pairs of legs in XX time. If you only see one pulse in XX time count as a pair of legs (one person). If you see three pulses in XX time (two people crowded in same doorway) count as two pairs of legs etcetera etcetera.
 
Milldrone

I can see that counting feet would increase accuracy, but setting up a light beam at ground level is prone to even more errors, with people kicking the beam/reflector, but I could perhaps set up 3-4 PIR sensors on a row from the ceiling each detecting a small ground area
 
PeterH_DK said:
Milldrone

I can see that counting feet would increase accuracy,



Then what about the one legged man or the long skirted woman or the high skipping child
errors are inevitable
even 2 people counting for 2 hours will disagree

find what you think will be a good reliable location and live with it or spend extra on video and 2 people counting again you will get differences

My experience is 2 shifts doing 5000 items with counter being reconciled against bar code readings of incoming items one shift comes in very close to counter seldom agrees 2nd shift is always less by a greater amount we live with it
 
A shopping mall i do work for has traffic counters on each and every door. Theses counters can tell if a person is leaving or entering the mall..they use these to figure out the high traffic areas and raise the rents in those areas. they also give the mall the daily stats, hour by hour. these people spent hundreds of thousands on these but you will never get a 100% accurate count..in never equals out...for example..i routinly go into the mall thru the door and then duck into a service corridor, from there you can get to any area and any store without being in the mall, it also takes you to the loading dock area and another large hotel,i also work at the hotel so i get counted going in but never out, This may happen 10-15 times a day..and thats just one person, there are thousands of delivery's per day..

when you say 3%....how many people are we talking about?
 
YCL: Thanks that is very interesting, will contact them on Monday



Darrenj and GIL47:

Thanks to both of you too, somehow or other I have expressed myself badly. I would be very happy with an accuracy of 3 % or maybe even 5% or 10% the problem is I am not sure what my accuracy is. I have measured a difference of less than 3% between a light beam and a PIR sensor - based on approx 500 entries per day, but I don’t know the true amount of entries at all.



If I can prove that I only count say 80% of the entries then I know that I must multiply the counting by 1,25.




I have looked at Trafsys recommended by YCL, they don’t state accuracy but might offer a solution that I can adapt. I will look at that on Monday

Thanks for the info and help

/Peter
 

Similar Topics

What is the raw count of Allen Bradley Flex Analog Output Module 5094-OF8 raw count?
Replies
1
Views
106
I am working on a Markem X40 printer which uses NGPCL Commands and needs to send commands using a .net program. I need help with the following...
Replies
0
Views
76
Hi, I'm programming in RSLogix 500, and I'm wondering how I would program a Jog command that does not increase the encoder count. Basically we'd...
Replies
3
Views
280
Hi all, I am currently working on a software in Automation Studio that is supposed to detect when effort is inputted by a user. This is done via...
Replies
0
Views
573
Hi i have project on micrologix 1400 to display cycle time count on HMI. I am beginner. in project i have six input once all input get high it...
Replies
3
Views
797
Back
Top Bottom