Dropcam has pulled out all the stops with its new Dropcam Pro monitoring camera by vastly improving its clarity, control and, better yet, brains. Physical improvements account for a better picture than previously possible on a camera that could fit in your pocket, but it’s the touchscreen-optimized controls (such as pinch-to-zoom) and machine-learning-based activity recognition — powered by Dropcam’s massive cloud platform — that really steal the show.
Better hardware, better software, better experience
Admittedly, I haven’t used a cloud-connected home-monitoring/surveillance/however-you-want-to-use-it camera before, so perhaps I’m easily impressed. But it’s really easy to see where all the work went into the Dropcam Pro. As soon as I plugged in a test unit and began watching my cats mill about (OK, laze about) the fairly wide expanse of my living room and kitchen, I was struck by the clarity of the picture and the new 130-degree viewing angle.
Dropcam Co-founder and CEO Greg Duffy (who’s speaking at our Mobilize conference next week in San Francisco) says the new six-element, all-glass lens and a significantly larger light sensor have a lot to do with that. All told, the camera is two times clearer in full-light situations and seven times clearer at low light and at night. This while providing a range of vision that’s about akin to a human’s line of sight, only there’s no periphery — everything is equally clear.
“You’re setting it in the corner of the room and you still want to see everything,” Duffy said, so Dropcam didn’t skimp on the premium parts. (At $199, the Dropcam Pro is $50 more than Dropcam HD, the company’s previous top-of-the-line model, which is now just called “Dropcam.”)
Here’s a video of a black cat licking itself in my living room in pitch blackness.
But the new components also make some of Dropcam’s software improvements even more impressive. People viewing their Dropcam feed — live or archived — via their mobile device can now use the classic touchscreen pinch gesture to zoom in on the areas they want to see. (Both Dropcam Pro and previous generations support the feature, although the former can handle 8x digital zoom versus 4x for the latter.) As with any other camera, though, the picture does tend to get fuzzy the closer you zoom.
But, Duffy showed me during a demo, Dropcam Pro’s larger light sensor — which is twice the size of the one in the iPhone 5s (and of the previous-generation Dropcam), he bragged — helps resolve this problem. When you zoom in on your device’s screen and press the little magic wand button, the camera takes in more light to that area and only streams that portion of the picture, resulting in a pretty clear image even of something as small as, say, an alarm clock that accounts for only a tiny fraction of the real estate in that 130-degree landscape.
“It makes us feel like we are living in CSI or Hawaii 5-0 or something,” Duffy said. And, he added, “It’s all about the putting the most light onto the sensor that you can.”
If you’re into specs, Dropcam Pro also sports dual-band Wi-Fi and a Bluetooth LE radio, which makes it easier to set up and will also let it connect with whatever other smart devices the Internet of Things brings our way and Dropcam thinks make sense to integrate with.
Bring on the machine learning
But hardware is replicable and, with good enough engineers and programmers, so is a similar user experience. That’s why the most-impressive thing about Dropcam Pro might be a new feature called Activity Recognition, which is available in beta (only via the web, for now, with mobile support coming) to customers of Dropcam’s cloud-archiving and video-management service. It uses machine learning to categorize the motion the camera detects and — ideally — make users’ lives a lot easier.
A camera facing a window, for example, might always pick up a tree blowing in the wind, Duffy explained. (Or, in my house, a camera in my entryway would always capture my dog moving back and forth between sleeping spots.) With Activity Recognition, Dropcam’s system will watch and learn for a day, and then begin putting events into categories, getting smarter the more it sees and the more the user tells it what’s what. At that point, users can remove certain events from their timeline, disable alerts for certain events (they do get kind of old) or just have an easier time scrolling their timeline to find what they’re looking for.
Duffy said Dropcam has been beefing up its computer vision team for the past 18 months and plans to keep advancing its work in this field to keep improving the intelligence of the cloud service, which is a big part of the company’s business. When I asked if Dropcam will ever enable keyword searches of video (e.g., Dropcam will automatically tag images its system recognizes), Duffy said it’s definitely a possibility. He has instructed his team of accomplished researchers to go wild figuring out ways to let users analyze all the video they’re uploading and find out the stuff they want.
“It’s basically all we think about,” he said.
In many ways, Dropcam is facing the same problem a company like Google is facing with YouTube only, well, with even more video content. At some point, you need advanced computer vision capabilities in order to recognize what’s happening in videos. Otherwise, every user’s archive becomes a vast, unsearchable wasteland and a company like Dropcam is wasting untold volumes of potentially valuable data in aggregate.
Dropcam has plenty of imagery on which it can train its computer-vision systems. Entry-level cloud plans store seven days worth of video, and Duffy said about 40 percent of Dropcam users are on some sort of video plan. He says it’s “the largest inbound streaming service on the entire internet.”
And, he added, “Dropcam at its core is a cloud services company that happens to make hardware.” The camera — even with all its high-end hardware — is just the delivery vector.