Now that we have sensors in some recycling centers, here are some observations about measurements and trajectories.
They could be used in order to create an algorithm in the sensor improving the precision of its measurements.
Disclaimer: Theses are only assumptions, and may not apply to every recycling centers or even be relevant at all.
Let's take a look at some results.
Here is the affluency graph for Gensac, the 22/12/2015:
It seems to be pretty accurate. We don't even have any devices detected when the recycling center (RC, in short) is closed (12h30 to 13h30).
Here's another affluency graph, but for Sauveterre de Guyenne, also the 22/12/2015:
This one is not bad, but we still see some device(s) when the recycling center is closed... That's not good. We'll take a look at it.
Analysis with 6stats
Trajectory lifetime
Let's see what happened for Sauveterre de Guyenne.
We see that there is only one signal between 11h and 12h50. (when it's supposed to be close), but tweaking the signal strength filter and the standard deviation filter is not really useful here. Let's see the trajectories:
Okay, the device we "see" when the RC is closed stayed here for a while. We suppose it's the guy who works here.
We can then filter every trajectories lasting more than 2 hours, for example, and, yep, its better !
If we take a look at Gensac, it could also enhance our measurements. Here, we filtered trajectories in order to only see thoses who stayed for 2h or more.
Even though they are not visible between 10h30 and 12h30 (which made it harder to detect), There are some devices (Mostly the dark blue and green ones.) that we measure and are probably irrelevant (here too, it's probably the guy (or guys) who works at the RC).
So... Here's a first idea:
Maybe the sensor should fliter MAC addresses that it regularly sees for a long time.
Signal strength
Note : This one is very RC dependant
I'm not sure if this one can be useful, but if we take a look at some graphs (Gensac too, and it seems to be recurrent)
It looks like we can make a difference between devices that are really close from the sensor (probably in the shed), and the others.
It needs to be confirmed, and is not 100% accurate, since we also see trajectories with a low lifetime above the red line.
MAC randomisation
We could only observe this one in the Node, and not in any RC, but it could be a clue for this 6sense issue?
I couldn't take a screenshot for this one, but there are many trajectories made of a single point, always in the same signal strength range (very close).
IMHO, it may be a iphone randomizing its MAC address. If it is the case, it's a good thing, because it does not change its MAC address more than once in 5-10 minutes, and shouldn't cause too much problems with our way to measure devices.
Some measurements analysis
Now that we have sensors in some recycling centers, here are some observations about measurements and trajectories.
They could be used in order to create an algorithm in the sensor improving the precision of its measurements.
Disclaimer: Theses are only assumptions, and may not apply to every recycling centers or even be relevant at all.
Let's take a look at some results.
Here is the affluency graph for Gensac, the 22/12/2015:
It seems to be pretty accurate. We don't even have any devices detected when the recycling center (RC, in short) is closed (12h30 to 13h30).
Here's another affluency graph, but for Sauveterre de Guyenne, also the 22/12/2015:
This one is not bad, but we still see some device(s) when the recycling center is closed... That's not good. We'll take a look at it.
Analysis with 6stats
Trajectory lifetime
Let's see what happened for Sauveterre de Guyenne.
We see that there is only one signal between 11h and 12h50. (when it's supposed to be close), but tweaking the signal strength filter and the standard deviation filter is not really useful here. Let's see the trajectories:
Okay, the device we "see" when the RC is closed stayed here for a while. We suppose it's the guy who works here.
We can then filter every trajectories lasting more than 2 hours, for example, and, yep, its better !
If we take a look at Gensac, it could also enhance our measurements. Here, we filtered trajectories in order to only see thoses who stayed for 2h or more.
Even though they are not visible between 10h30 and 12h30 (which made it harder to detect), There are some devices (Mostly the dark blue and green ones.) that we measure and are probably irrelevant (here too, it's probably the guy (or guys) who works at the RC).
So... Here's a first idea:
I'm not sure if this one can be useful, but if we take a look at some graphs (Gensac too, and it seems to be recurrent)
It looks like we can make a difference between devices that are really close from the sensor (probably in the shed), and the others.
It needs to be confirmed, and is not 100% accurate, since we also see trajectories with a low lifetime above the red line.
MAC randomisation
We could only observe this one in the Node, and not in any RC, but it could be a clue for this 6sense issue?
I couldn't take a screenshot for this one, but there are many trajectories made of a single point, always in the same signal strength range (very close).
IMHO, it may be a iphone randomizing its MAC address. If it is the case, it's a good thing, because it does not change its MAC address more than once in 5-10 minutes, and shouldn't cause too much problems with our way to measure devices.