After that, pretty cool idea and nice site!
The one thing that really confuses me is the pricing structure. Why do you need to pay each month? Paying $300 for one year of usage seems pretty steep.
It's service pricing but it seems like one-time service/purchase.
To be a member of the queue of an establishment can carry some of the social status of that establishment itself. This is a primative notion yes, but fairly ingrained culturally. As businesses often judge their prestige by the length of the queue to enter.
What really matters, in my opinion, is whether having a long line during certain peak times generates more revenue for the business (by being able to charge more, or by gaining attention due to its long queues) or whether providing a faster and better customer experience with shorter wait times does (and perhaps more customers during non-peak times)
Whether one is better than the other may depend completely on your business. Also, I'm not sure this technology forces you to publish your current density information. You could simply use the information for business intelligence gathering and for providing promotions during non-peak times.
We've considered things like Helium. Eventually centralizing the network will be good but for now, wifi's cheap and available.
Could it be used for tracking population leaving during a fire drill? As often this is the first question asked by first responders, "Is anyone still inside?"
imo.
We just do installs at each entrance and then normalize.
How does the device tell how many people are present based on counting at a doorframe? Is it counting people passing through the door? Does it detect direction? Is it good enough to estimate the number of people inside by subtracting out counts from in counts? Will this work for wide entrances, such as malls? Is this a passive infrared sensor? Those go blind in hot weather.
This looks like a cheaper alternative to video counting systems, of which there are many. Video systems get about 98% count accuracy. If you already have surveillance cameras, you can often use them for counting. Beyond that, there's queue measurement - not just how many people are in line, but how many gave up and left without buying.[1] (Seven people in a queue is the tipping point – any longer and most shoppers won’t bother joining it. After 9 minutes, shoppers are likely to give up queuing and leave empty handed (other research says as little 6 minutes). 70% of customers who leave never come back.)
> which implies they're selling data to someone else. Who?
Typically our customers are startups who sell to SMBs (coffee shops, bars, restaurants, museums, etc). They charge merchants anywhere between $50-$500/mo/location for some kind of software or service. These are startups that sell POS systems, loyalty software, marketing services, discounts, handle logistics, and delivery.
> There's no privacy policy. No terms of service.
Frankly, it should have been there before launch but since people don't "buy" through our website, we decided to sacrifice legal thoroughness for speed to launch. Maybe a misstep but people seemed okay emailing us their request to order.
> How does the device tell how many people are present based on counting at a doorframe?
Two closely situated, parallel infrared distance sensors. We timestamp spikes in voltage as they come in allowing us to see o...1 = entrance. 1...0 = exit. Giving us the current count in a place.
> Is it counting people passing through the door? Direction?
Yes. Not the line outside. Although we can do line detection and estimate wait times. Yes.
> Is it good enough to estimate the number of people inside by subtracting out counts from in counts?
Yes. It's better than just an estimate.
> Will this work for wide entrances, such as malls? Is this a passive infrared sensor?
No. Our current model maxes out at roughly 90in -- that's with two sensors on either side of a double door facing one another. No it's AIR.
> 98% count accuracy ... If you already have surveillance cameras, you can often use them for counting.
You're right. We're just betting that customer-aversion to facial recognition and surveillance cameras is slowing adoption in the long tail of the market we're after - the various independent merchants and sellers that comprise a city and who are too busy making coffee and food to spend too much time on potential controversial technology. See: http://techcrunch.com/2014/05/29/philz-coffee-drops-euclid-a...
[edit: for readability]
> Yes. It's better than just an estimate
What if someone waved their hand in front of it (in a particular direction) in an attempt to subvert it? Couldn't someone trick it into over-reporting or under-reporting the number of people inside?
What if someone is standing idle in front of it (near the door frame) for a few minutes and in the meantime a large number of other people walk in? Will it miss them?
What if two or more people walked in at the same time with no gap between them?
Most importantly: How would the counter ever be able to correct itself from inaccuracies?
Also I don't think this data is particularly attractive to anyone except competitors... and even then its not like its an email list or anything, merely traffic.
Might I add that if there's no privacy polity, the FTC won't go after you for violating it. And the company is new, so this stuff is "tomorrow" kind of stuff. They have the limelight for right now; might as well use it.
> How does the device tell how many people are present based on counting at a doorframe? Is it counting people passing through the door? Does it detect direction? Is it good enough to estimate the number of people inside by subtracting out counts from in counts?
Doesn't seem so. Looks like they're going for the niche between mom-and-pop to something smaller than Staples.
> Will this work for wide entrances, such as malls? Is this a passive infrared sensor? Those go blind in hot weather.
PIR doesn't die due to hot weather. They go blind in heavy humidity. They start throwing errors and bounces around 40% and up. Here in Indiana, they would be nigh useless right now, unless inside AC.
> This looks like a cheaper alternative to video counting systems, of which there are many. Video systems get about 98% count accuracy.
I'm working on that. I've gotten better than 98% already. My time is "make it better and sell it".
> If you already have surveillance cameras, you can often use them for counting.
That's my point. Most/all businesses already have it. So the whole creep and privacy factor is a moot point. Might as well leverage the tech you already have installed.
The sensor wires in to an electrical outlet. Andrew specified that the wire is custom length; indicating it is manually handled at install.
As far as I can tell from the updated website, the new sensor is the same thing, just dressed up nicer. Each sensor is only counting one metric, i.e. ingress. Beyond that, the base hardware (not the sensor) is a Raspberry Pi that connects to both the sensor(s) and WiFi. I'd imagine it would make sense to have many sensors for each box, but maybe not.
I wish however, you'd include a section that provides a reasonable explanation as to why Workroom got a 950% increase in traffic. I'm thinking that if I saw that a place was full, I'd be less likely to head over there. But, if I didn't know, I'd probably give it a whirl. Alternatively, if I've gone there a bunch of times and it's always been full, I just wouldn't go. What exactly is happening?
Also though, I think it was a 950% increase in traffic to the web pages powered by this product, not to the actual locations.
When I use google maps, it shows me how busy various roads are and it also chooses the fastest route based on how fluid the traffic is. Seems like google maps is just observing the world and making decisions based on those observations.
Now I was wondering, what if all the drivers used google maps at the same time ?
Wouldn't it mean that google maps is influencing and even creating the traffic patterns ?
Same here - just measuring the 'density' has the effect of actually influencing it which is an interesting outcome and resembles quantum mechanics voodoo stuff :).
I don't want to be taken away from your page, I want to continue reading more about YOUR product before I go to view your "partner" sites. Same goes for your API, Twitter, etc.
> Why does this need to connect to your servers? Because that's where the signal gets processed. We store all signal data ever captured and do all the math on an analyze server. The benefit of this is that as our hardware can be cheap enough to replace or give away and when our algorithm improves so too does all historical data.
> What happens if you become insolvent? Not sure yet. What happens if any of your providers become insolvent?
> Why can't I just buy the damn thing then use it? Because there's no margin or recurring value in the damn thing. And selling hardware is a bad idea right now.
> Why do I have to pay for a subscription to a service which is effectively a simple proxy between the device and my infrastructure? Because we're not selling the device. We're selling the analyzed data. Also, I should probably say, you're not our target customer. We're after companies that control / manage large networks of locations not individual developers.
To be honest, we'd love to give this to any developer who wants it and just see what they build. Hopefully, we'll be able to do that someday and support that community properly. Until then, this is the best way for us to be a sustainable business and not rely on the crazy thin and ever-decreasing margins of hardware sales.
Hope that addresses some of your questions.
Andrew
IMO, this is the strongest section: http://www.density.io/#comparison
After she got 9/10 correct (she's 58)... she looks at me and says, "Ship it, Andrew. Ship it."
You don't have to resort to infrared sensors in order to provide anonymity, but they do inherently limit you potential accuracy and metrics you can prodive. It all comes down to where the data is processed and what is made available. Also, cheap camera components are coming down fast due to economies of scale, not sure how IR will play out in the long haul.
[1] http://www.informationweek.com/strategic-cio/executive-insig...
Looks like you've got a JS issue in there somewhere.
- How well does it handle people walking in side by side?
- What about someone pushing a cart or stroller?
- Can it tell if someone is walking in or out?
- How far does the laser reach? Would I need two sensors if I had a double or triple wide door?
- I assume these are battery powered? How long do they last?
Really glad to see a product coming into the regular consumer space for this :)
His answers regarding can it see the traffic direction (answer - yes for each individual) and how it is powered (custom length power cable, cut at install time for both sensor and base) are inline in this thread.
Also Andrew explained the base gathers info from all the sensors for counts.
I asked also about people in a group - most doorways are a physical narrowing that automatically places people in single file. If the doorway is wider than that, they can place a sensor on either side (my assumption is that these results are then filtered so a person is not counted 2x).
Something the paper pointed out was that while the temperature from animal to animal varied, the same animal often kept the same temperature (+/- epsilon) when moving from sensor to sensor and that give some idea of uniqueness.
And while I also love kefka's solution I think the face recognition stuff gets a more Orwellian reaction than just tracking heat sources.
So, the second problem with cameras/face tracking is cost?
I wrote this: https://github.com/jwcrawley/uWho
It does what your hardware does, but also does facial recognition. It's still early in the build process, but I'm working on a commercial (non-QT) version of this.
But right now, it can accurately count unique people, as well as remember people. So when Jane walks in front of the camera, it remembers that her database number is 1234. Tomorrow, it will also remember that she's 1234. And tracking is all stored in few XML config files for easy calculation.
So you're using the slippery slope as justification and welcome sign instead of warning. The world is all downhill from here exactly because people don't have the integrity and courage to resist.
Also
I can hire 2 guards who can remember people by taking a photo. And then I can have them recall who shows up. With people.
All my project does is substitute a computer for human. The only reason why we don't do the above is because people cost a lot more. Computer software and cycles are cheap.
And the procedure I used with my code saves a hash of the face. I cannot generate faces from the hash, although it would be a one liner to spool a face to the hard drive when a pic is captured. My software doesn't do that.
In the interests of science, why did you switch from WiFi signals to infra-red door counters? Was it that you picked up passersby? (See [1] for previous pitch, which explains it more than the archive.org website)
[0] https://web.archive.org/web/20140605031145/http://www.densit...
If this interests you:
This type of thing can be done at the hobby level. Since this is mostly an IR sensor with wifi connectivity talking to an API and sending mqtt data.
http://sensalytics.net - Deutch project counting people, mobile tracking and cash register data.
http://i-counter.ru - Russian projects - uses IR counter with 3G transmission module, so data send directly to web analytics server without additional networking devices.
Sometimes google maps will detect heavy traffic on I-70 and start re-routing drivers the other way... unfortunately this very quickly creates a bottleneck that Google Maps can't detect in time (traffic goes from 0mph to 60mph to 0mph in the mountains) so it'll continue to funnel people down that "shortcut" until the traffic essentially equalizes with the I-70 traffic.
There's an even worse side effect, as those who went through Idaho Springs eventually have to get back onto I-70 to get to the ski resorts, so now that on ramp (which is a metered on ramp) backs up, further hurting both I-70 and the "shortcut" traffic. It's a real terrible feedback loop that essentially is caused by Google Maps not being able to adequately predict how much traffic the Idaho Springs route can handle, which seems like a hard problem to solve (especially generally).
Edit: Thought about this more and realized predicting ski traffic is more or less a proxy for predicting the weather, so I highly doubt this is a fixable problem (at least in this specific case)
The city could include data about road closures, police & fire incidents, etc.
This might even be used to reduce traffic along bus routes, thereby increasing efficiency for them as well.
If someone works in the T&LC and wants to implement this feel free. ;)
EDIT: Note this works particularly well in Manhattan where the grid system means there are generally many routes of equal length between any two points.
I'm pretty sure I've observed this in the wild. Occasionally there's a major accident on a freeway in Denver that totally stops traffic. People that know the area well enough hop off the first exit they can and find the most straightforward way over to a major arterial road running in the same direction. Maps sometimes comes up with the theoretically-fastest route that requires lots of turns through neighborhoods, so you'll see a whole conga line of cars piling through suburban streets and congesting the whole area.
This article is about Waze, but I can remember reading the same thing about google maps:
http://www.latimes.com/local/california/la-me-california-com... (Ctrl-f "variety")
I wondered what happens when you enter some sort of harmonic oscillation of people trying to evade traffic with alternate routes matching the delay of Google re-routing them.
The easy fix would i guess be doing a round-robin of alternate routes..
http://fortune.com/2015/05/22/meet-density-a-startup-that-le...
The cynic in me sees this as as a requirement to justify their (current or future) venture funding. So that they can justify a high valuation in their eventual acquisition and shutdown in the next 12-36 months.
It's not enough to 'just' build a good product and sell it to customers anymore.
Camera: Real-time, Accurate
Break Beam: Anonymous, Accurate
Density: Real-time, Anonymous, Accurate, CheapThere's absolutely no information on what they consider "accurate," and anything I can find on older HN threads is very vague and implies that accuracy wasn't their main focus.
Here's a basic question I have: can this count a mob of 20 people walking out of store at the same time? If it can't, I believe "accuracy" should have some context, since there are other enterprise systems that do that no problem.
Edit I am an employee of Density and definitely biased.
Placemeter does much more than pay lip service to privacy. We pride ourself on our privacy efforts. If you want to lear more about them, @afar email me: david@placemeter.com
It is just basic marketing and public relations. Don't try to sell people on your competition being lesser, even if it's true (which it is apparently not, in this case). That does absolutely nothing to sell me on your product, but instead makes me acutely aware that you see them as a threat which needs to be quashed.
For what it's worth, I also found your use of the word "homie" rather unprofessional, off-putting and patronizing, especially in a thread which acts as public announcement of your company's product.
These things combined have opened the door for your competitor to defend themselves in a more logical and professional manner, which I would say they've accomplished in this thread. I'm entirely unimpressed with how you are representing your company, but I am very impressed with how your competitor has responded.
Also, it would be nice if you would identify yourself as an employee of Density when you post comments.
The landing page advertises this as an entry-level system for doorframes. It doesn't seem that this is made for costco-level swarms of people entering a large space at one time. That said, I also have my doubts on accuracy but the error margin is probably acceptable enough to most.
And also, CCTV isn't open to the public or traded around wanton. Instead, I would argue, this data is highly confidential to the business and therefore would guard it selfishly.
And as for the pub photographing: There has been a trend in bars in Indianapolis to scan he barcode on the back of the drivers license to "verify identity". What they're really doing is building up a clientèle database for which they can do whatever with. And they are booming bars.
A classic example would be if a road is severely congested, and satnavs started advising drivers to take an alternative, smaller road, if too many people get the same advice that road will then end up even more congested and delay the people who followed the advice to change routes even more.
http://www.jstor.org/stable/42747332?seq=1#page_scan_tab_con... is a good starting point if you are interested in reading more about this area.
A common example I've seen used is: Automatic headlights on a car turning on in the dark. When the headlights turn on when you're in a garage or tunnel, they can create enough ambient light to trick the sensor into thinking it's light, and turning the headlights off. The solution here is to add a delay into the reaction.
I'm wondering what a solution is to a problem with mapping - is it add a delay to re-routing traffic? Or rerouting traffic randomly? Or spread traffic out over a number of routes? It's made harder by the fact that Google in this case won't be routing all traffic - just a subset.
Feel free to connect me with the right people, my email is in my profile
https://support.google.com/gmm/answer/6194894?hl=en
----
Your directions were wrong
At this time, you can’t report wrong turns from your phone or tablet, but you can report them on your computer.
Open Google Maps on your computer. Click Directions directions button image. Enter the starting point and destination for the route for which your directions were wrong. In the bottom right of the map, click Report a problem.
----
I'm working on getting a human contact involved.
There's a built-in bypass in the site.js to allow the context menu if shift is held, though you still get the alert. Or it's just that holding shift allows the context menu to appear after the alert. I don't actually know.
document.body.oncontextmenu = function() {
alert('Please dont right click!!')
return window.event.shiftKey // top secret
}
Why would someone want to know this information? Ideas abound! I don't think we've had such easy measurement and monitoring at the fingertips of the general populace.
Having detailed logs of user growth is invaluable (the much-vaunted hockey stick), however this level of granularity is unavailable in the physical world.
Now we can show interested parties that, for example, our storefronts are hockey-sticking, growth is high and projected to grow higher, etc. You get the idea.
The alternative to this platform are little devices that monitor things like cellphones that walk into the room, and use this to approximate occupancy.
Other products like this are quite expensive to implement and are geared more towards large scale deployments, not small businesses. I even looked into building my own sensor for this, but I simply don't have time. Happy to see this product emerge.
My 2nd favorite: 950% of 0 is 0.
My other question is how you derive count without uniquely identifying someone. If I'm entering a shop and a PM sensor sees me, will it know when I leave?
1) the processing is happening aboard the sensor, counts are what are sent back to the servers
2) we don't do unique identification like that. We use object detection, which is different than using unique biometric markers like face detection. That means that we can track a person or a car within a frame of view, but not if they exit and re-enter the frame like in the case you described.
Does Density still use wifi pinging for part of its counts?
No wifi pinging. After Apple almost killed us a year ago with their MAC address policy change and we realized there was significant push back on privacy, we dropped the technology altogether.
Yeah, isn't it terrible that there are companies trying to get money out of customers... :P
My comment is tongue-in-cheek but the macro point here is why should a company that invents something or provide a value need to do so in a way that minimizes their profit margin?
Product of tiredness and being trigger-happy with the reply button today.
Apologies to dfine and Placemeter.
Anonymity is a more interesting problem than "can distance sensors uniquely identify a passerby among 500M North Americans" as your reply implies. What if you are the roundest or tiniest person in town? All the sudden you are uniquely identifiable.
You're early in development, and I bet whoever did the circumference estimation has more in mind. I imagine you could make a good estimation of a person's height from your data: whether the profile sees knees hips or hands. Can you identify the asymmetric waist bulge of a CHL carrier? Does your infrared band penetrate polyesters but not cotton? I hope the reader's feature-vector blood is flowing at this point.
Your page needs to have way more formalization around the concept of anonymity (outside of the registration required area) for me to feel that you are appreciating the problem from an engineering perspective and not a marketing one.
Left wondering why you need a base controller and the dual-PIR's though, especially if the PIR's are connected to DC voltage anyway and not battery-powered like some other commentor said.
I get you would wanna use wifi to talk up to your cloud and maybe that's what the base is for. But seems like a waste if you got something expensive like a Raspberry Pi or other ARM A-series in there. There are much cheaper wifi chipsets out there now these days that could do the same thing without needing all that cost and space of an ARM Cortex A-core, plus wifi, Linux, etc. You could probably get away with putting a small wifi chipset in each of the dual-PIR housings and still come out ahead in cost. The ESP8266 sounds perfect for it actually.
Either way, a pretty cool product.
It can determine traffic direction; which is a leap beyond current technology i.e. a break beam, which cannot.
all it would take is two or more beams and some pretty simple code, am I missing something?