Density Debuts New Way To Count, Report And Display Crowds And Lines

August 10, 2016 by Dave Haynes


There’s a podcast episode up today about Density, a San Francisco start-up that has launched a new type of people-counting sensor I think has some interesting potential for the digital signage market.

Density was launched by Apple alums and former founders of Y Combinator. Their product is touted as a simple sensor-as-a-service that uses infrared beams to count people and estimate things like line-up lengths and wait times in venues.

The company’s sensors are discreet–about the size of an Apple TV–and made to fit above most doorways. Using sensors and the company’s API, Density users get access to depth sensing technology, machine learning, and computer vision to derive accurate people counts without any of the privacy concerns that come up – fairly or not – with camera-based or smartphone-sniffing  systems.

Here’s a partial transcript of the podcast interview with CEO Andrew Farah …

Dave: Andrew, tell me what Density’s all about.

Andrew: Well, we build a small sensor that monitors how busy a location is in real time. We do it anonymously. We really wanted to see how busy our favorite coffee shop was, a couple of years back, actually. There really wasn’t any inexpensive way to get access to an API for how many people had wandered in. We started to build a sensor, and today we’re about to launch a sensor that is our first mass market product, but it allows us to anonymously measure plus one and minus one into a location.

Dave: When you say “we,” who is we? There’s obviously a company behind this? How large a company, and where did it come from?

Andrew: Well, we’re about 16 people now. It actually started as a spin-off from an R and D studio that I founded with a number of my friends in college. We would fund different product development that we thought was interesting with fee for service work. Any time we’d find a large company that would want to build a web application or a mobile app, and we would build that for them, and then we would use that capital to invest in product development. Density was the seventh product, and we thought it had enough legs that we spun it off as a separate company, raised capital for it, and then transitioned the whole team over. We’ve been together for about five years, although Density as an entity has been around for about two and a half.

Dave: How does the product work?

Andrew: Our newest sensor gets affixed. It’s self-installed above a door. It illuminates a space with infrared light, and that infrared light bounces off the floor, and it comes back, and it has a certain amount of time that it takes to return back to the sensor. There’s a differential when someone walks through. The amount of time it takes for light to bounce off of someone’s shoulders is less than off the floor, and that difference is a depth value. We are able to measure tens of thousands of depth values every frame, at 30 frames a second or greater, and can as a result see these blobs of light or silhouettes as they walk through. We don’t actually know who you are, or what your ethnicity, gender, or age is. We’re able to run computer vision against these depth values, or these silhouettes, to figure out directionality, or if you’re colliding with someone, or even if you’re holding hands, we can distinguish between two people. It’s really amazing, the output data.

Dave: This is a lot more than just a … I forget what they call the things, but trip meters, basically, that break a light beam and count, “Okay, somebody broke the light beam. Somebody else broke the light beam. Now there’s twenty people who broke the light beam going through the door.”

Andrew: That’s correct. This is a networked sensor, so you connect it to a location’s wi-fi, or to a power over ethernet, and it’s a powered sensor, so it requires power as well. What’s cool is, you can have multiple entrances to one location, and Density, as an infrastructure, will reconcile the correct count for that one room, even if there are multiple places that people are entering and exiting simultaneously. You can do that for any number of rooms or entrances, and you can even query which entrance or exit are people most likely to go out of? When is our busiest time?

I should say there’s two really important things to know about density. One is that we build a device and an API. It’s incumbent on our customers to integrate people count into whatever system they care about most. We think that people count is most relevant in the context of other data, or other applications, and not necessarily as a one size fits most dashboard. The other thing is we don’t sell the hardware. We sell access to the data on a recurring basis for a low monthly fee, or a low annual fee. What that allows us to do is retain ownership of the hardware such that if the technology gets better, we can replace that for free. You have no interruption of service.

Dave: It also means your footprint just grows, and grows, and grows regardless of whether a client drops off, right?

Andrew: That’s correct, yes. You picked up on something that I think most people miss. If a customer churns, there’s a lifetime value associated with the physical sensor. The asset. We can re-deploy that sensor to a new customer.

Dave: You’re also aggregating a whole bunch of data that, if you can narrow it down to things like coffee shops, or bars, or hair salons or whatever it may be, you can start to develop much more global patterns that are marketable, right?

Andrew: Yes. That’s true, although one of the things that we learned very quickly was that people count may be a bit more of a universal problem than we initially expected. We’ve heard from a lot of coffee shops, and bars, and restaurants, but we’ve also heard from a lot of loyalty applications, and point of sales providers, who have networks. Very large networks, tens of thousands of small businesses that they work with. Those are cool, but what’s also cool are the financial institutions that want to do corporate space visualization, or the homeless shelter network of 70 plus homeless shelters, where they said, “This is the first time the technology’s been affordable enough for us to be able to count the number of homeless we serve, and then apply for bigger grants so that we can serve more people.”

We’ve heard from people we didn’t expect. Everything from tech startups, to realtors, to insurance companies and stadiums, airlines, indie developers. We’ve heard from hospices and hospitals, ERs, and what’s really exciting about that is that while we think aggregate level data is really interesting, we also realize that if you deploy a sensor in a network and you want access to that data so that you can solve a problem like staffing, or overcrowding, or lines, or something, you can do that within that network. You don’t need to make your data publicly accessible. Sorry, that was a bit of a monologue, but …

Dave: You’re infrastructure. You’re not providing a final product that’s going to give you lovely, rich color heat maps and bar charts and everything else. You’re providing an API of data, and it’s incumbent on whoever wants to use it to apply it in the way they want to use it, even if it’s just real time data that’s going to trigger something.

Andrew: That’s right. That was actually a huge internal debate, was, “Do we build a dashboard, or do we not build a dashboard and sort of have restraint, and see what customers build?” We ended up on the latter for a variety of reasons, but you’re absolutely right. We’re the infrastructure. We do the same thing, no matter the space. Our customers integrate that data into the software that matters most to them.


Dave: You’re in your own way potentially disrupting some other technologies that are being applied to digital signage, or there’s companies trying to do it, like facial pattern detection, video analytics, or sensors they’ve gotten. There’s those wi-fi sniffing, on and on and on, all these different things that are trying to establish what’s actually going on in an environment, so why would they use what you have as opposed to what these other companies offer?

Andrew: Well, we really don’t focus on … I mean, retail back in the 1980s was one of the earliest places where people came to matter. Ever since then, people analytics, and capturing demographic data, has been a real focus for that particular industry. The retail industry. The problem is that the technology got really expensive, and it started to gather more and more information about consumers, because the budgets that were paying for the technology were advertising dollars, or marketing dollars. It’s perfectly reasonable to say, “We just want to know more about our customers.” The unintended consequence, however, is that that technology could never break free of retail in any meaningful way.

There are a lot of other places where people counters get deployed, but by and large, there really hasn’t been a system that’s democratized people count, and then made that really easy and modern to access. The disrupting other technologies, I think if we were to go into retail and really try to spend all of our time there, that might be a bit more in line, but our intention actually isn’t to compete with any of those groups. It’s to make it possible for a library, and for a non-profit, and for the City of San Francisco, and transit systems, and airlines to adopt technology that was otherwise unattainable for a handful of reasons. It was either too invasive of privacy, it was too expensive, or it wasn’t modern enough to access quickly.

Dave: Let’s talk about costs in a broad sense. If I am a library, and I want to understand things about what’s going on in my library with my members, my guests, and so on, what would be installed? What are they looking at in terms of costs, and how would that compare roughly to if you’re using other kinds of technology?

Andrew: Sure. We’re pretty close to launch, but I would say that pricing is always a heated debate. I’ll speak to the prices that we currently have. They’re kind of subject to change a little bit, but I think that these are the ones that are final. We have two plans. You can either buy the sensor for $50 per sensor per month, if you pay annually, or you can buy the data at a month to month no term for $95 per sensor per month, which means you can just return it after 30 days and only have spent $95. We get a lot of requests for, like, volume discounts, but we are trying to be really fair with our pricing. We think that this is a reasonable enough fee that we can reduce, if the term grows really long.

Density is most relevant if you have it for a long time. You can start to do some really interesting predictive data. You can do some really interesting historical data. The longer that gets, the more valuable that history becomes. The other thing is that it’s also most valuable at whatever scale the company operates. If I’m a major theme park, for instance, getting 25 sensors isn’t going to matter much. I need 5,000 for it to matter. That’s very exciting for us, because, you know, we’ve got a lot of different types of customers from every different industry. We’re excited to see it start as a side project, and as sort of an internal thing that they’re testing, and then turn into infrastructure, to use your term.

Dave: Let’s talk about something like a big amusement park, a theme park, or a big event center like a stadium. Like T-Mobile Arena in Las Vegas, which just opened. That sort of thing. How would you deploy there, and what would it be giving the end user? Andrew: Well, in a stadium circumstance, there’s a couple of different uses that we’ve heard from NFL teams and NBA teams. The obvious one is, “How long is the bathroom line? I’d really like to tell the fans.” The beer line typically comes next. Then, then it starts to get into …

Dave: They’re all pretty connected.

Andrew: Yes. That is so interesting. Once you unlock the data, you can start to predict when the bathroom might actually be overcrowded. I’ve never thought of that. That’s a good idea. The other thing is that it starts to move into operations facilities. People are saying, “Okay, where are the most used spaces, and do I need to deploy cleaners there?” Or, “We’ve got a secure area. Are we seeing a lot of traffic through there that we shouldn’t be?” Or, just general human load balancing. “I’ve got a lot of people in this area. I’ve got very few people over here. How can we kind of change that?”

I think you’d also said theme parks. A lot of the data that is currently available is either, “I gathered the data because I have a surveillance camera and I’m doing some really advanced computer vision on top of that,” but you sort of run into privacy issues, but that data is really relevant if you want to do any kind of incentive based balancing. If I’ve got a lot of people in an attraction that is performing really well, and others aren’t, being able to measure that over time might be able to effect whether or not I bring in that attraction or expand my investment on that attraction. I’m not really sure. Theme parks aren’t a big industry for us, but I do think that there are sort of two uses. There’s internal, which is the business itself, and then there’s end users. Consumers like you and me, who might be attending these locations.

Dave: From the lens of digital signage, you were talking about informing customers. Let’s go back to something like the washroom lines and the beer lines, and so on. You’ve got sensors that are steadily scanning an area, and measuring the numbers of people there and so on. That’s spitting out data that in theory, a content management system would be able to steadily query and have rules that would then say, “All right. It looks like the waiting period here is five minutes, but I know that 100 yards down, at the stadium concourse, it’s wide open, or it’s a one minute wait, or so on.” That’s something that could then be displayed on a screen with wait times, and saying, you know, “Wait time is only one minute 100 yards this way.”

Andrew: That’s an interesting way to bring it back to digital signage, is that … Also, just to clarify, our sensors don’t scan … First off, they’re self-installable, so someone can just, like a Nest. You would get it in a box, and you install it above a door, or above an entryway. The other thing is, we don’t scan a room. Instead, the illumination goes directly down. The infrared illumination goes directly down, and as people walk through that entryway, we can only see about four feet into a space, and about eight feet wide. We support any standard single or double entryway.

To your point about signage, yeah, the data can live anywhere. I think that’s probably the most exciting thing about it, is that instead of it going into a dashboard where you need to use it only if you’re a facilities manager, and looking at historical trends, and it’s like, Google Analytics for the real world, which isn’t terribly interesting for the rest of us, I think having an iPad, or a sign up, or some kind of screen showing you relevant information about the space, and where you might be able to go, like, “The fan store right now has no line.” It might be really busy where you are right now, where you’re reading this information, but if you know that you can buy the baseball that you wanted to get and you don’t have to wait in line, that might drive more sales.

Dave: Is the data that you could get through the API already kind of structured and everything else so that a developer working for a content management system company, or let’s say an interactive agency or something like that, could work with that data pretty readily? Or is it a big project to get integrated with Density?

Andrew: It is ten calls. It is the most straightforward API we’ve ever worked with. You have count. You can pass us a time stamp, and a location ID. You could give us multiple location IDs, and we simply return for you based on whatever those parameters were, on how many people were in that room at what time. Eventually, we’ll add abstractions on top of that, so that it makes it easier, where you can say, “Give me the peak time for these 25 locations.” We can kick that back without you having to do any computation. There’s some really easy, sort of low hanging fruit, API improvements. You can actually go to and it’s all right there. It’s really straightforward. We’ve tried to make it as easy as possible to interact with a sensor.

Dave: A developer looking at this will kind of go, “Okay. All right. Fine. Great. No problem”?

Andrew: That’s right. In fact, we’ve even put the docs in front of a lot of the developers that have contacted us over the last year, and asked for their feedback, and then iterated based on that feedback, and overwhelmingly, it’s like, “Okay. This is super explicit. Feels like Twilio. Really easy to use.” That’s exciting for us, because functionally, that’s our product. People have to get at that data as easily as possible.

Dave: Is anybody else doing this? Are you competing with other companies?

Andrew: It’s a good question. I think it depends on the lens you look at it. There are companies that do thermal, they do stereoscopic vision. There are groups that do video cameras, and facial recognition. There’s beacons, and all sorts of other things like NFC, or RFID I mean. I think there are very few that have taken a business to consumer approach to this, and thought about this as not about the hardware. It’s about the data. We’re going to abstract away the cost of the hardware, and instead just charge you for what we believe the value of the data to be, and incur the rather considerable cost of the hardware ourselves, just to make your experience super easy. Then, the other thing is, the physical design of this product is built so that it can go into or complement any space. I think for something that can live in a space for five plus years, it should be something that isn’t an eyesore. I think that unfortunately, there are a lot of companies that I would say fall into the category of people counting, but when you really look at the products side by side, the differences are so much more significant than the similarities.

You can listen to the full podcast, or 16 earlier chats, here …

Leave a comment