Artificial Intelligence. You’ve heard the words in just about every facet of our lives, just two words, and they’re quite possibly the most moving, life-changing words employed in everyday conversations. So what exactly is AI, who currently uses it and should be using it?
What is AI?
AI is a powerful way of collecting, qualifying and quantifying data toward a meaningful conclusion to help us reach decisions more quickly or automate processes which could be considered mundane or repetitive. AI in its previous state was known as “machine learning” or “machine processing” which has evolved into “deep learning” or, here in the present, Artificial Intelligence.
AI as it applies to the security and surveillance industry provides us the ability to discover and process meaningful information more quickly than at any other time in modern history. Flashback – VCR tapes, blurred images, fast-forward, rewind and repeat. This process became digital, though continued to be very time-consuming. Today’s surveillance video management systems have automated many of these processes with features like “museum search” seeking an object removed from a camera view or “motion detection” to create alerts when objects move through a selected viewpoint. These features are often confused with AI, and are really supportive analytics of the Artificial Intelligence, not AI themselves.
Fully appreciating AI means employment of a machine or series of machines to collect, process and produce information obtained from basic video features or analytics. What the machines learn depends on what is asked of them. The truth is, the only way the AI can become meaningful is if there is enough information learned to provide the results desired. If there isn’t enough info, then we must dig deeper for information or learn more, properly described as “deep-learning” AI. Translated, this means that we need to learn more on a deeper level in order to obtain the collaborative combined information necessary to produce the desired result.
Deep learning AI
Deep learning AI can afford us the ability to understand more about person characteristic traits & behaviors. Applying this information can then further be applied to understand how to interpret patterns of behavior with the end goal of predictable behavior. This prediction requires some degree of human interpretation so that we are able to position ourselves to disrupt patterns of negative behavior or simply look for persons of interest based on these patterns of behavior. These same patterns evolve into intelligence which over time increases the machine’s ability to more accurately predict patterns that could allow for actions to be taken as a result. This intelligence which is now actionable could translate to life safety such as stopping a production manufacturing process, if a person were to move into an area where they shouldn’t be which might put them in danger.
Useful applications of intelligence
Informative knowledge or intelligence gathered could be useful in retail applications as well by simply collecting traffic patterns as patrons enter a showroom. This is often displayed in the form of heat mapping of the most commonly traveled paths or determining choke points that detract from a shopper’s experience within the retail establishment. It could also mean relocating signage to more heavily traveled foot-paths to gain the highest possible exposure to communicating a sale or similar notice, perhaps lending itself to driving higher interest to a sale or product capability. Some of this signage or direction could even translate to increased revenues by realigning the customer engagement and purchasing points.
From a surveillance perspective, AI could be retranslated to actionable intelligence by providing behavioral data to allow law enforcement to engage individuals with malicious intent earlier, thus preventing crimes in whole or in part based on previously learned data. The data collection points now begin to depart from a more benign, passive role into an actionable role. As a result, new questions are being asked regarding the cameras intended purpose or role of its viewpoint such as detection, observation, recognition or identification.
Detecting human presence
By way of example, a camera or data collector may need to detect human presence, as well as positively identify who the person is. So the analytic trip line is crossed or motion box activated or counter-flow is detected which then creates an alert for a guard or observer to take action. Further up the food chain, a supervisor is also notified and the facial characteristics are captured. These remain camera analytics, but now we feed this collected facial information to a graphic processing unit (GPU) which could be employed to compare captured characteristics with pre-loaded facial characteristics. When the two sources are compared and a match produced, an alert could be generated which results in an intervention or other similar action with the effort of preventing a further action. This process- detect, disrupt, deter or detain could be considered life-saving by predictably displaying possible outcomes in advance of the intended actions.
The next level is deep-learning AI which employs the same characteristics to determine where else within the CCTV ecosystem the individual may have been previously by comparatively analyzing other collected video data. This becomes deep-learning AI when the GPU machine is able to learn from user-tagged positive identification, which the machine learns and begins to further reprocess its own data to further understand where else the person of interest (POI) may have existed on the ecosystem and more correctly improve its own predictive capabilities, thus becoming faster at displaying alerts and better at the discovery of previously archived video data.
In conclusion, the future of these “predictables” wholly rests in the hands of the purchasing end-user. Our job is to help everyone understand the capabilities and theirs is to continue to make the investment so that the research perpetuates upon itself. Just think where we’d be if purchasers didn’t invest in the smartphone?