[OpenTRV-interest] Fwd: Count crowds with Bluetooth Connectivity Kit
mail at mikestirling.co.uk
Wed Dec 30 09:32:02 GMT 2015
On 30/12/15 09:13, Simon Hobson wrote:
> Nigel <nigel at discreetsecuritysolutions.com> wrote:
>> Counting people though? Bad idea - without any other way to check it, it's going to basically be a random number.
>> So no: you can't get even a rough guess on crowd numbers with Bluetooth without loads of extra data.
> I think your conclusion comes from incorrect assumptions.
> Yes, for small numbers you are correct. In (say) a group of 10 people, some will not have taken their phone with them, some will have BT turned off, and you really wouldn't know.
> But for a large crowd, and bear in mind that they are talking about 10s of thousands here, it becomes statistically valid to say that "on average X% of the population carry a phone with BT turned on, we've counted Y BT devices, so there are very roughly about Y/X people here". The key thing is knowing X - I don't know it, but I bet there are experts in the field of counting people who do have a good idea. People have been counting people for longer than I've been on this earth - it wouldn't be hard to run old and new alongside each other to empirically determine a value for X under different circumstances.
But how many people would have BT turned on _and_ advertising all the
time (else it's as good as turned off from a detection PoV)? These days
I'd say that number is close to 0 - certainly Android only advertises
for a limited period of time (older versions) or while the BT settings
menu is open (later versions). This is as much a battery saving
exercise as it is privacy protection.
> So if you are in the business of counting people, I don't think X is hard to determine (and it'll vary by situation and demographic). Once you know X, you just knock up a device to count Y and you've got a pretty good idea of total numbers. Not "there were 54,157" people there" accuracy, but at least "there are 50-55k people there" levels of accuracy which is probably more than accurate enough for this sort of situation.
> BTW - for counting devices you don't look at the device name, you'd look at the MAC address which, if the manufacturer isn't a complete idiot, is globally unique.
For Bluetooth Smart (BLE) it is more likely you'd be able to detect
devices as they tend to be advertising most of the time; they can do so
with a far lower power budget than BT Classic. For example, a year or
so ago when I started doing some BLE work in the office there were no
devices other than my own, but now I see Fitbits, Apple watches, etc.
presumably from people in the offices downstairs. The problem here
though in relation to MAC addresses is that for BLE they are often
More information about the OpenTRV-interest