The Marriage of Artificial Intelligence and Video Analytics
Traditional CCTV Integrators May Be Missing The Boat!
After spending the last six plus years selling and installing integration of video surveillance systems that deliver what is obviously needed, I’ve come to the realization that security integrators, in general, are not grasping “the big picture”.
As a guest speaker at a California Alarm Association meeting in San Diego some years ago, I stressed to the members that they needed to “up their game”. It seemed as if their main focus at the time was selling and servicing smaller CCTV systems, most of which were going into residential homes. Very few were selling to the larger commercial accounts. Even those that were servicing the larger commercial market were not able, or were hesitant to begin the integration of computer-controlled systems and video technology.
My experience over the last five plus years in the security system integration has shown me that there is still a vast difference between what these integrators are pushing versus what is available and is actually happening on the technology front. Some companies in the IT industry are seizing this huge opportunity, while the CCTV / security industry system integrators do not seem to be technically adept enough to pursue these solutions. If they don’t, I believe that they will cede this mainstay of their industry to IT. They seem to be more focused on capturing client dollars with long term support contracts because their margins are so thin, whereas if they could invest in higher level technology integration, margins would explode.
The integration of Artificial Intelligence with Video Analytics is a rapidly growing market. Not only is this marriage allowing CCTV operators to focus on only important tasks, it is growing in the area of predictive analytics. AI can help anticipate potential future scenarios utilizing self-teaching capability. Using much of the same technology as weather predictions, AI can provide assessments of what might happen next.
Additionally, AI addresses a key weakness in traditional security CCTV monitoring: Human boredom!
The human mind evolved to keep humans alive. Our brains are keyed to visual cues, and very quickly notice changes. However, they are not that good at watching an unending scene without a drop in alertness. It is estimated that the human brain only remembers 25% of what is observed when a scene does not have any appreciative changes occurring. Subconsciously, the brain “turns off” the “alert” mode when nothing is happening. This leads to lapses in concentration on what is being observed. Especially slow changes are rarely noticed at all.
Enter Artificial Intelligence!!
An AI system only has one mode when it’s used to monitor video, and that is “Full Alert”! Not only can it detect minute changes, it is observing 100% of the time, even when nothing is happening.
AI can take over monitoring and only point an operator’s attention to important situations. This allows operators to focus more on dealing with only actionable items. This is a major manpower issue in the central station monitoring centers – personnel only have a certain amount of time they can spend observing any given view, and then they have to move to a different view in the effort to keep an eye on everything.
Additionally, and obviously, AI can also be utilized when doing searches of recorded video, which is a manually intensive task. An AI system can instantaneously bring up video that has specific characteristics desired by the searchers.
For instance, you can instruct an AI system to search for any woman wearing a blue dress. This task could take hours using traditional searching methods, but only takes a few seconds when utilizing an AI/CCTV integrated system.
This is enabling security analytics to be performed in real-time in the typical areas where CCTV has been utilized: Monitoring electrical power plants, nuclear research facilities, warehouse/ distribution centers, retail stores and outlets, automotive and manufacturing lines, and onshore and offshore oil platforms to name just a few.
For example, in the area of manufacturing: Analysis can be performed on various processes, and redundant or a potentially dangerous motion can be detected quickly versus hours of human analysis.
Much has been said of facial recognition, but there is much more advancement in the individual identification area than most people realize. For instance, a persons’ gait can be identified and assigned specifically to them. This can enable the tracking a someone who dons a disguise to be tracked through a crowded airport.
Yesterday in a San Francisco parking structure, I sat in my car waiting for a phone call. I observed a man looking around while he was walking from one car to the next, carrying a duffel bag and looking in cars with a flashlight. It seemed as if he was searching for valuables, and so I alerted management. They immediately went into action, informed security and called the police, as there had been a number of break-ins of parked cars.
An AI/CCTV integrated system could have immediately spotted this type of activity utilizing predictive analytics and reported it much faster than I did. It could even have made a prediction about where this person might go, where he entered the parking structure, and track him if he tried to evade security.
The applications of AI/CCTV integrated systems can be expanded into so many areas that its use is still unrealized. Traffic analysis and public transportation utilization can be performed. Retail stores can quickly assess which displays are attracting the most attention to optimize sales. These same systems can also simultaneously look for shoplifting activity and alert the appropriate personnel.
When and AI/CCTV system has a Pan/Tilt/Zoom camera, it can be programmed to track an individual through a store, in a parking lot, or track a vehicle through traffic and pass the image on to another camera when the vehicle leaves the area of view.
As Artificial Intelligence utilizes neural networks and is set up to “self-teach”, it can analyze video and come to conclusions, often times in mere seconds, about what is important and what is not. Referencing that retail example above, person walking normally through a store and leaving might not create any significant data. But if a person starts stopping at certain displays, and then makes a purchase based upon what they’ve seen, the AI/CCTV system can follow that person and put a pointer on that segment of video for analysis by marketing. Combining that data with a close up to see their face could allow for micro expression analysis, potentially revealing their reactions to a specific display. Repeat that several hundred times a week and a retailer can glean some very exact data about what makes a customer purchase a product.
Just imagine how effective this would be in a supermarket, or a new car lot, or an electronics store. It’s already being done with online shopping, but there is a huge market for people shopping out in the real world.
The applications for AI/CCTV systems are just beginning to be realized, and its capabilities are barely being tapped. With the cost of CCTV systems going down rapidly, and AI systems growing in capability almost exponentially, we are just now entering the “honeymoon” stage. It seems that this marriage might just be at the beginning of a long and happy relationship.
About the Author:
Rick Downey spent over twenty years working in High Performance Computing with the leading innovator in computer graphics and visualization, and an additional six years in the video surveillance industry. These two diverse career paths have given him a unique perspective on the merging of Artificial Intelligence and Video Analytics. He has provided integrated visualization solutions in complex environments, serving various commercial, state, local, federal and educational customers nation-wide working for a variety of companies including Oracle, Johnson Controls, SGI and Cray. His home is in the Santa Cruz redwood mountains adjacent to Silicon Valley.