Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: Density – Anonymous People Counter and API (density.io)
258 points by jordanmessina on July 13, 2015 | hide | past | favorite | 145 comments

The site is rather vague. They provide the hardware and install it for free, which implies they're selling data to someone else. Who? There's no privacy policy. No terms of service. The "order" button just brings up a blank email.

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.)

[1] http://www.retailsensing.com/queue-management.html

[Density founder here] Hey Animats, sorry it took me all day to get to your list of questions.

> 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]

>> 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

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?

Hardware/install is free, but costs you $25/month/location to receive your data. (Not that this means they are not also selling data elsewhere).

Yup, as soon as I read 'free hardware, free to install' I immediately though 'Service model'

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.

I would say it's extremely useful -- most establishments have a general idea of foot traffic (ie: POS transactions or qualitative visual assessment) whereas with density.io you have a detailed history of when and where there's foot traffic.

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.

Oh the data itself is interesting, but not particularly sell-able to others, in my opinion.

The alternative to this platform are little devices that monitor things like cellphones that walk into the room, and use this to approximate occupancy.

It is. If you had a competitor who got most of their business around 8pm on Tuesday. You could offer huge savings at 8pm on Tuesdays driving away traffic from their business.

I own a retail store and have been wanting this type of data to quantify behavior, find correlations and discover patterns. In it's simplest form, it could help to quantify previously 'untrackable' advertising methods.

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.

> The site is rather vague. They provide the hardware and install it for free, which implies they're selling data to someone else. Who? There's no privacy policy. No terms of service. The "order" button just brings up a blank email.

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.

Also, what kind of battery does it take and how often am I going to need to change it? Honestly, maybe it's because I have too many devices but I'm done with anything which requires a battery swap more than a once every couple years.

Andrew @ Density.io answered this for me.

The sensor wires in to an electrical outlet. Andrew specified that the wire is custom length; indicating it is manually handled at install.

I was also wondering about its ability to determine the direction of traffic. If it relies on a location having a separate entrance and exit, that's a pretty big negative. I'm also wondering how it handles large groups. Does anyone have any idea how the hardware works?

There's not much actual information available (this should be concerning), but there have been some clues dropped into past HN threads. This is a picture posted in another HN thread of their "original sensor" which is just 2 basic IR sensors: https://s3.amazonaws.com/screenshots.angel.co/73/340163/7982...

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.

Hi, that's indeed our old sensor which is composed of two AIR sensors. The new sensor carries a slight upgrade in hardware, but the real power comes from our signal processing. See my reply to thwest here. https://news.ycombinator.com/item?id=9879671

So you use 2 different IR sensors, which "measure your circumference to some degree" and provide a "distance profile" which somehow translates into real-time accurate foot traffic counts. IMO that's definitely a cop-out explanation, and per usual, there is not any actual quantification.

Hmm, look's like Vishay's VCNL3020 in there? Or the good ol' Sharp GP2Y0A21YK0F.

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.

Using an Rpi gets you to market much, much faster. Then optimize for cost after it gets popular.

Andrew @ density.io answered this for me.

It can determine traffic direction; which is a leap beyond current technology i.e. a break beam, which cannot.

> 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?

Nope, not missing anything. There are 2 beams (IR) per one of their "sensors", and it's assumed they're determining direction by noting which beam fires first vs second.

Fantastic product idea and a landing page that's so good that I figured out what the product does without touching my scroll wheel.

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?

I think the idea would be that this helps fill dead times. If the place is full, they don't need/want more people showing up just to be turned away. If the place is empty, then they want to advertise that to fill it back up.

Also though, I think it was a 950% increase in traffic to the web pages powered by this product, not to the actual locations.

Right, the increase was on the webpages that included Density to show seating capacity.

I also find this particularly confusing/misleading - since you deal mainly with physical locations (often called sites by owners), using the word "site" instead of "web traffic" seems odd. Also, listing a web based metric as large as 950% without any type of time interval really kills the true significance of that metric.

My favorite statistic: 73.6% of all statistics are made up.

My 2nd favorite: 950% of 0 is 0.

Oh, that wasn't clear at all.

Yeah. We could and will clear that up.

Is that sustained website growth since then? Also can you make that distinction clearer on the site? It's definitely misleading some people.

>> "After adding Density, we saw as much as a 950% increase in site traffic to supported locations. Our users love it." - Darren Buckner, Workfrom CEO

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 live in Boulder and this happens all the time during ski season. When traveling towards I-70 there's a longer, more circuitous route which breaks off of Highway 6 and goes through Idaho Springs. This path is longer, but cuts off the beginning of the initial drive into the Front Range... it's also just a two lane mountain road.

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)

Would you be willing to report this to the Maps team? If not, just reply and I'll get in touch with someone about it.

Can't seem to find where to file this bug since it's not a physical mapping error. Coincidentally the Maps team is in the Boulder office, are they not? I guess I could just walk down and say hi :)

Feel free to connect me with the right people, my email is in my profile

Try this:



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.

Couldn't it take a guess based on previous actions? If it does this every weekend and keeps getting the same results, ease up a bit? If it tracks individuals, it should be able to tell when it see it diverted someone, only to have them end up in traffic.

Since in NYC taxis are a substantial portion of road traffic, I once had the idea that NYC should have a special navigation platform (maybe based on Google Maps?) that the taxis are mandated (or, more likely, highly encouraged) to follow, which would essentially load balance the traffic.

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.

> Wouldn't it mean that google maps is influencing and even creating the traffic patterns ?

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.

It'll be interesting to see if Google starts to proactively load balance these sorts of situations - send people on several different routes at once to keep traffic flowing.

This problem is an active area of academic research in the transport operations/planning field. It's generally known as "concentration and overreaction".

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.

It's similar to the 'flip-flopping' problem (I'm sure there's a technical term for it) - actions based on feedback from a sensor changing the sensor reading, triggering a different reaction.

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.

Waze does this, and Google owns Waze.

There is a policy on the backend of google maps to diversify the best routes they suggest to users. It's like A/B testing where they show different responses to users who make the same requests

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 totally had that same thought when using google maps to navigate LA.

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..

Yeah it seems like at some point you'll end up in a traffic "bubble" where you don't actually get the "fastest" route but the one that thinks will make the roads fastest for everyone.

Yeah... we're secretly hoping for some version of https://en.wikipedia.org/wiki/Schrödinger's_cat

I thought Google Maps used the Android phone location to know the busy road?

And anyone actively using maps.

Please don't be too clever for your own good - let me Cmd-Click on the links to open in a new tab.

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.

This is obnoxious website behavior, and I equate it to sites ignorantly disabling right-clicking because they're paranoid someone will steal their photos. If your site contains any links at all, you should never, ever be breaking such basic browser functionality. The only possible legitimate use case would be a webapp which is attempting to create a desktop-style environment, and even then, co-opting the CTRL or ALT keys is likely not the best course of action.

I disable right click on my Website.

I'll shift+right click all I want on your site. ;)

For what reason? This is easily bypassed.

If you visit his website [0] it's (at least I hope) intentionally poor designed, everything terrible straight from the late 90's in web design. From marquee, blinking, embedded mp3, and more.

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
[0] bitfission.com

That is an amazing website, including the MP3-rendered MIDI, and I'm now pretty sure the comment was a joke that everyone missed :)

We agree. That was not by design. Updating.

Why does this need to connect to your servers? What happens if you become insolvent? Why can't I just buy the damn thing then use it? Why do I have to pay for a subscription to a service which is effectively a simple proxy between the device and my infrastructure?

[a Density founder here] Okay...

> 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.


"Density has raised an undisclosed amount of seed funding from Ludlow Ventures, Jason Calacanis and Bill Draper. "


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.

That is my biggest thought. If this is oriented to hackers, let me own the thing so I can do some really cool stuff with it.

I hate this user hostile trend among hardware startups, that is requiring/involving third party "cloud" infrastructure for no particular reason other than to extort money out of customers.

for no particular reason other than to extort money out of customers

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?

The hardware sounds dead simple, not to mention cheap. The service is the actual product here.

I don't think the hardware is simple. How would you design something like this? I've thought about this problem for years and no elegant solution has occured until I saw this.

Hmm? Two beams detecting directionality + WiFi?

Hardware's intentionally simple. We've even considered open sourcing it :).. [Density founder here]

Had no idea what this was. Clicked link. Within one minute I figured out what Density is. Nice job on the landing page!

IMO, this is the strongest section: http://www.density.io/#comparison

At one point we went as far as to quiz our moms. "Mom, what does Density do? What does it cost? Do we sell surveillance cameras?"

After she got 9/10 correct (she's 58)... she looks at me and says, "Ship it, Andrew. Ship it."

"Hacker News Tested, Mom Approved".

Don't camera systems beat this in pretty much every way? http://www.placemeter.com/ comes to mind.

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.

From the landing page:

  Camera: Real-time, Accurate
  Break Beam: Anonymous, Accurate
  Density: Real-time, Anonymous, Accurate, Cheap

Yeah, I find that claim particularly amusing since they're just using a single PIR sensor, which is far less accurate than any camera based system. There's no way that you can claim a single PIR beam is as accurate as a camera based system.

There'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.

> "With Density on a doorframe, we’ll tell you."

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.

Most camera systems wouldn't accurately count the mob of 20 (I work on camera count analytics, and we have to correct for this)

Placemeter looks very smart although they have an enterprise sales model, Density is literally giving their units away...

Yeah... I mean I appreciate that people pay lip-service to privacy but a surveillance camera that counts you as a distinct individual without your consent is an invasion of privacy. The cost is higher than the price tag, homie.

Edit I am an employee of Density and definitely biased.

I am a PM at Placemeter. We do not "use facial recognition software" in our algorithms, as Density's website claims of video-based systems. None of our algorithms use biometric markers for our counting—we're essentially the same "dumb" counters as Density's IR with the added advantage of accuracy and area of coverage.

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

I am curious. Particularly with your new sensor. It sounded like much of the video processing happens on the unit itself, meaning faces never reach Placemeter servers. Is it accurate to say that you're only getting counts and movement data?

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?

Great questions.

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?

That's really interesting. Given a certain level of granularity, would it be possible for a person to have a unique object signature? I guess at that point, you'd just use a face. Just curious.

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.

Not quite sure I understand your question. We store counts, not individual object IDs, so at an individual granularity it would be the same as your IR device counting one person.

I think you should probably readjust your tact for any more comments today and in the future. Making insults, implications or even allegations about a competitor is never going to be in your company's best interest.

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.

You're right. It was an unprofessional comment. I was flippant and insulting. You are also certainly right to be disappointed, off-put, and impressed with how PM responded.

Product of tiredness and being trigger-happy with the reply button today.

Apologies to dfine and Placemeter.

It happens. And I didn't mean to publicly shame you, either. There just happens to not be private messages here, so public was the only option.

No worries. I totally earned that one.

If a camera films you while you are on someone else's property, then I'd argue it isn't an invasion of privacy since there was no expectation of privacy to begin with.

Also, it would be nice if you would identify yourself as an employee of Density when you post comments.

I agree. But I think there should be a distinction between being on someone's property and being recorded for security and being in a cafe, having your face recognized as someone that frequented a sex shop last week, pairing that with an online profile, and then being pushed a retargeted ad campaign about your fetish.

For the record, that's not what Placemeter does. We don't individually identify people with our algorithms. (I am a PM at Placemeter.)

What the....?

Any plans to offer a "disconnected" version where I could push this data to my own aggregator+API instance? Use case for me would be where foot traffic numbers would itself be sensitive data.

This isn't in the pipeline but shoot us an email and maybe we can work something out: team@density.io

Our student consulting group, Optimir, worked with Density to test their product around UC Berkeley's campus. It's a fantastic product, and we're really excited to get these installed permanently in libraries, coffeeshops, and weight rooms around campus.

(some) Kroger chain stores have a similar technology [1] to this to determine how many cashiers are needed at any given time! Neat stuff!

[1] http://www.informationweek.com/strategic-cio/executive-insig...

Bug report: In Firefox, when on this page [1], clicking on the Density logo in the top left takes me to http://www.density.io/undefined

Looks like you've got a JS issue in there somewhere.

[1] http://www.density.io/?ref=hn

Thank you! Fixing it now.

I have a few questions for the Density folks:

- 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?

I have always wanted to build something like this since I was a kid. The biggest driving thought was to make traffic signals dynamic and build a central communication system for a city's transport.

Really glad to see a product coming into the regular consumer space for this :)

I spoke with Andrew over at density.io.

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).

Does this use PIR out of curiosity, or are you using an emitter, receiver in one package.

We’re using infrared sensors. Both the emitter and detector are on a single sensor, so we only affix our hardware to one side of a doorframe. This is different than break beam which requires hardware on each side of a door.

How much accuracy do you lose when people are dressed up in full winter gear?

When you would wear metamaterial stealth suits maybe you can avoid being detected by our IR sensors, we're curious as to what you have available!

Incredibly accurate and anonymous seem to be at tension here. A discussion on your signal processing would be illuminating.

We made our own AIR door counter, composed of a few distance sensors. It can tell us in which direction a person is passing, and gives a distance profile that we process to distinguish individual people. Technically, we measure your circumference to some degree, but you might agree that is far from enough to compromise your anonimity.

AIR being active infrared as opposed to passive (or a particular waveband)?

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.

Gotcha, yeah that makes sense, as it'd cut down the number of individual packages you'd need for the system.

Looks great! It could be very nice with public transportation, so I'll be able to know beforehand which wagons are not overcrowded.

This is pretty awesome. I mentioned doing something like this at the IoT talk I gave, I had read a paper on counting animals by their heat signature for conservation purposes and thought something along those lines would be a simple replacement for the 'walk through the door ring a bell' detectors, and given that a 32bit ARM Cortex-M is about $1.25 you could do that for not much more money than the old light + photocell.

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.

Ok. So businesses already have cameras. Yeah, privacy, schmiravacy. That horse has done gone left the barn.

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.

> Yeah, privacy, schmiravacy. That horse has done gone left the barn.

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.


Slippery slope? Humans already can do classification. Should we ban humans from identifying and counting other humans?

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.

You see, humans have judgement, and obvious presence. I don't mind George and Jim remembering I was at the pub on Tuesday, because they will consider who to reveal that information to before they do. (A pub I don't think anyone would be at if there were two guards photographing everyone). These giant piles of databases are open for everyone.

Doesn't the business have a right to keep track of its clientèle? Obvious legalities of underage-ness of a pub aside, I already can hire counters that watch the security feed and assess numbers.

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.

This reminds me of Motionloft. Their CEO was an idiot and ran into some legal trouble [1] which hurt the company, but it was backed by Mark Cuban and is still running afaik. I have thought since first learning about Motionloft that this could be a big business but I don't understand why both companies refuse to just sell the hardware. There have to be a ton of companies that would buy it but don't want their very sensitive data about customer counts etc. being transferred to a third party that could be hacked etc.

[1] http://techcrunch.com/2015/04/01/cuzyolo/

Soo I thought I had seen this somewhere, just with WiFi used to measure users, turns out it comes straight from the past![0]

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...

[1] https://www.youtube.com/watch?v=yRGa9-QUDWo

Something like one of those points rewards cards with rfid and this could allow a merchant to get your purchase data as you enter the premises. Then with other rfid readers throughout the store you can pinpoint where $user spends most time and adjust $coupon for them to incite purchase. Or maybe multiple IR beams like this one in different parts of the store to have traffic metrics on the floor plan. Looks very nice from the design standpoint. It is something that will not register in the mind of the customer.

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.

There are already solutions that do this using your phone's wifi: https://www.washingtonpost.com/blogs/the-switch/wp/2013/10/1...

Great product and great team. Congrats everyone!

How does your product compare to existing traffic solutions, like Shoppertrak?

Cheap and anonymous by design. Also, we're not after big retail. We're after everyone else.

There many similar startups over the world:

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.

Looks like an awesome product and hardware but the SAAS model sucks (for us). Here's to hoping they release a LAN model for a one time fee.

Your home page freezes my Firefox (33.1 on Mac) :( . Have to pound through five or six unresponsive script warnings.

After that, pretty cool idea and nice site!

Sorry. If you can email team AT density.io with a bug report or something that shows the latency that would be really helpful.

This seems really cool! I can think of a bunch of new things to do with this and a lot of places where I want to know how busy they are.

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.

This looks awesome! Would love to know ahead of time how crowded my local coffee shop is before I packed up to work there.

If this is the indented use case what purpose the coffee shop have to add one? If in most cases it'll only discourage customer attendance.

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.

Too long of a queue could discourage customer attendance as well, as people may not want to wait a long time, no matter how prestigious the business is. In addition, a lesser-known business could use lower wait times as a competitive advantage.

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.

How do these talk with the outside world? Do you need to attach them to your own internal WiFi?

Yeah. Currently we piggy-back on a locations wifi. Relatively low upstream data (37kb / 90secs).

We've considered things like Helium. Eventually centralizing the network will be good but for now, wifi's cheap and available.

Gorgeous web design, kudos.

Thanks. Took a long time to get the right balance of imagery, icons, and copy.

Does anyone know why is this product "Show HN" while the other product currently on the front page, Soloshot, only has a product name and slogan?

I did the Show HN. The original title was "Density Platform" because we just revealed our new sensors and the public facing API. Not sure why a mod changed it.

There will be a big business putting these on all the entrances of a publicly-traded casino and then tracking activity levels.

What communications protocol is it using?

We're using infrared sensors to detect ingresses and egresses from locations.

How does the solution handle multi-ingress/egress points?

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?"

So that's a really novel idea. It would be incredible to tell a first responder how many people we think are inside... accurately.

We just do installs at each entrance and then normalize.

This would be a reallly cool application if PIR's were capable of counting "horde" foot traffic, like a giant mass of people streaming out of a door during an alarm. I don't think that something like a PIR could handle that number of people, since it's just the single beam.

I really like you, for some reason. I think it's your green handle.

Thats not a protocol. Is it subgigahertz with Simplicity, or Zigbee or ...etc, etc.

Looking forward to hearing more about it when you're ready to share

Could this be done with Apple's iBeacon as well?

It can! To accomplish this there'd need to be a custom iPhone app for a location and everyone that goes in would have to have the application installed and have their bluetooth turned on.

Foot traffic should be inexpensive, accurate, and available in real-time. It should be dead simple to integrate with other applications, and it should never invade people's privacy.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact