Deep Sentinel mixes manpower with artificial intelligence
The development of Deep Learning computer algorithms in conjunction with the proliferation of high-resolution cameras have helped revolutionize the video surveillance industry. The capabilities these respective technologies deliver have given rise to a new generation of video analytics technology, commonly referred to as artificial intelligence or AI, which are helping end-users make better use of both the security and business operations data these systems are generating on a daily basis.
But while intelligent video analysis tools are helping to reduce the burdens produced by false alarms and sift through video footage faster, some believe that the technology isn’t being leveraged to its full potential. One firm that hopes to transform how AI is used within the industry is Deep Sentinel, which combines both analytics software and human guards to create a comprehensive video monitoring solution for both residential and commercial security applications.
The company was founded by David Selinger, an early employee at Amazon who also co-founded real-estate brokerage firm Redfin, who was motivated to address what he saw as the inability of technology alone to prevent crime after a neighbor of his suffered a home invasion.
“I was meeting with the police and our neighborhood watch and [our neighbor] had a really extensive, like $10,000 [video surveillance] setup and a burglar alarm, but as I was asking the police with all this, this level of investment, why didn’t it stop the crime. The answer was embarrassingly obvious, which was ‘what did you expect that stuff to do?’ Selinger says. “That was my aha moment that this isn’t about drawing a red box around that and saying there is a person at your front door or even that there may be a person breaking in, it is about building a business process that is vertically integrated between what technology can do and what people do with it.”
Combining People and Technology
After contemplating this industry disconnect, Selinger says he realized that he would have to build out his own monitoring center and employ a roster of guards who could interpret the data generated by cameras leveraging AI software.
“As a company, we focus on the AI and software that connect cameras and security equipment to the guards,” he explains. “It is not a horizontal [sale] where we provide boxes and AI in that sense, but we think about our end-to-end loop as staring with the cameras and ending with the actions and behaviors of the guards and what makes that different is we don’t think of the guards as the end-user and that an important distinction. We think of the guards as our primary source of data, meaning when a guard clicks on a video to review it and thinks that it is interesting, that’s actually the most interesting piece of data that is generated – period.”
For example, every time a guard decides to intervene or call the police regarding an event that has been captured by the cameras, Selinger says all of these “ground truths” can be tied back to the Deep Learning system using weights and variables in the video – i.e., was there a crime committed or something suspicious that needed more investigation – to improve the performance of the AI. Deep Sentinel also tracks each time a guard clicks on a video stream to view it, which enables them to assign severity to event.
“The very first piece of software that we wrote was a guard console. We don’t use the standard 4x2, 2x2 or 3x2 grid console. We wrote one that has actions and data annotations integrated into it, meaning that when our guards sit down at their console, the events come into a specialized console that is designed to track (incidents) and then enable them to respond quickly,” Selinger adds. “The reason we have our own console is because you can either invest a ton of time and training to have elite trained guards or you can move as much of that (data) into the software itself. We’re working hard to make that console the place where all of that risk goes.”
When an event happens Selinger says they use on-premises AI which can do three thing initially: identify what is in the field of view, if it is a person then determine if they are doing something that might be concerning (i.e. entering an area of interest, picking up a package, etc.), and finally it determines whether or not it should be seen by a guard. If the AI determines a guard should view the event, it then decides what visuals the guard receives and provides them a list of actions that could be taken, such as using two-way audio to notify the possible intruder they are being recorded, call the property owner or the police should it be necessary.
Go-to-Market Strategy
Deep Sentinel, which counts Amazon Founder Jeff Bezos among its lead investors, is currently selling through the security integration channel to both the residential and commercial security markets.
“We believe that installers are an important part of this equation, so we sell through channel and installers are a key channel,” Selinger explains. “We do that for several reasons. One is PoE. After doing some tests early with our direct-to-consumer [sales] and letting them install their own [cameras], it turns out you can mess up a lot of stuff easily and they did. Secondly, this is a new product. There aren’t a lot of companies that have figured out how to do active guarding and there’s certainly not any that have figured out how to vertically integrate AI with guards, so we needed to make sure we had a trained set of partners. We have about 300 of them across the U.S. right now that don’t oversell the abilities… and make sure they are really familiar with how to do this and do it well, as well as support customers and bring feedback to us in an effective way.”
Though the company recently announced support for deployments of up to 20 cameras per property and also rolled out features tailored for specific vertical markets, such as car dealerships, warehouses and construction sites, Selinger emphasizes that this is still a solution that makes sense for many homeowners as their residential package begins at $100 a month for three cameras.
“Business is maybe 50% to 60% of our sales, but that means that the rest is residential, and we really crush it on residential,” he says. “By the way, it is not just high-end residential. Our mission is to stop the crimes and if you have someone that owns a home that is $200,000, but they are in a neighborhood where they are getting broken into a couple times a month or they just don’t feel safe – you don’t have an alternative. If you are in a situation where you don’t have thousands and thousands of dollars to spend, we’re at the price of a mobile phone plan.”
Joel Griffin is the Editor of SecurityInfoWatch.com and a veteran security journalist. You can reach him at [email protected].
