Advance or AI based video analytics can detect potential terrorist from a crowd of passengers at an airport terminal, or analytics will recognize a face of a wanted criminal in a jam-packed football stadium are all marketed by companies, racing to capture more customers.

How much of this is fact versus fiction is debatable, nevertheless many matured and practical analytics features without the lofty claims have vastly improved the effectiveness of video surveillance and security management.

Today, there are 125 companies offering VMS (Video Management System) and a host of analytics features that claims the use of artificial intelligence, neural networks, machine vision, self-learning, behavioral analytics and more.
The use of analytics is not limited to security & surveillance industry and within the security industry, requirements for Homeland Security are significantly different than Commercial Security. Trillions of data points are generated every minute across the globe and they can be manipulated and analyzed to generate beneficial statistical and other data for businesses and operations in various ways. The analyzed data can improve security, efficiency, target marketing, advancement in science, space, medical research and hundreds of other uses.

It is highly illogical to assume video analytics is a new invention. Long before the invent of IP cameras, the first generation of video analytics was using Codec that converted analog signal into digital pixelated signal to process the pixel changes that generated alarms like, motion detection. Conversely, the primary focus of early analytics was for indoor applications as outdoor environments were challenging to monitor for pixel movements due to natural factors including, sun, clouds and wind which can cause false alarms, as an example, a tree branch movement in windy conditions. Subsequent improvements in algorithm and processing power allowed developers to define the pixel changes as objects based on shapes, size, speed, aspect ratio and other factors to define them as cars, people, dog or a tree, so analytics use even in outdoor environment is possible, today.

Today, video analytics can raise alerts in real time, during post event review and define common denominators across multiple cameras during live or recorded sessions to find as an example, a suspected robber wearing a green color jacket who robbed a convenience store. This level of analytics is useful to track live events and to help with forensic analysis to reconstruct what has transpired during the course of the crime. The important thing to remember is, this scale of analytics is more useful for law enforcement and justifiable as compared to an office building environment, where the use of it may not be commercially appealing.

Building up on the earlier generation of analytics, the current analytic offerings can be extremely powerful as crime prevention and fighting tools, in addition to producing powerful business intelligence data. Using edge or processor based analytics they can be used to extract highly useful data to benefit many businesses and users from increased productivity in a factory to increased sales for a retailer. The retailer can easily align their operations based on the deep analysis an AI assisted BI(business intelligence) data can provide from occupancy levels, to peak times, gender information to approximate age and engagement level of customers.

The new generation of analytics is more purpose built, faster and accurate, but equally dependent on having adequate processing power to achieve desired results, that can vary from simple processing to more complex computations. As an example, an ANPR camera using advance algorithm integrated with OCR (Optical Character Recognition) and other characteristics can rapidly capture number plates of passing vehicles and also able to recognize a black listed vehicle. The processor requirement for recognizing a black listed license plate is significantly higher than just capturing a license plate.
In another example, if you are implementing a Facial Recognition system to identify known and unknown persons in a high security area, require careful study of not only the environment where facial recognition cameras will be installed but what are the realistic customer requirements? Do they need notification when an alarm is triggered that an unknown face is detected, or do they want to identify the person from a blacklist, in which case the facial recognition database require recognizable facial details of all blacklisted persons to trigger a positive match alert and necessitate additional processing power and more challenging algorithms.

Facial recognition is very taxing on any processor and is notoriously imperfect due to many dynamic conditions that can impact the recognition criteria. Many basic analytic features like tripwire, object detection, line crossing, masking, directional alarm, camera sabotage and people counting can be simultaneously used in a scene and dozens or even hundreds of these analytic cameras can be connected to a server depending on the processor specifications.
In case of ANPR and Facial Recognition analytics, the number of cameras that can be connected to a server can be significantly less as their respective algorithms will require considerably more CPU and GPU power to perform their assigned operations.

In order to achieve reliability, high accuracy and desired results, it is imperative to use best practices and manufacturer recommended guides for server specs, camera resolution, positioning, best lens and suitable lighting, but most importantly the customer expectations, commercial impact and realistic aim must be clearly understood and agreed with the customer.
Humans takes things for granted from recognizing colors to faces, however for analytics or even artificial intelligence, visual intelligence is still in its infancy and it is a “program” that is defined with fixed parameters with some variances and require a lot of learning, before it can detect anomalies and start to make decisions, on its own. While there are visible improvements in ”recognizing patterns” and “behavioral analytics” or self-learning analytics, the cost of implementation may not be commercially attractive yet, to use them commonly in commercial security applications.

What the new generation of analytic technology can deliver looks great, however it is equally vital to not over-promise as many analytic features are marketed without long term testing in actual environments and tend to fall short of customer expectations.
Every project will have its unique setting and the dynamic nature of actual environment as compared to a laboratory setting or controlled environment, therefore if proper pre-qualification for the suitability of a particular analytic feature is not done and expectations not managed, it can become a source of customer dissatisfaction.

In order to achieve reliability, high accuracy and desired results, it is imperative to use best practices and manufacturer recommended guides for server specs, camera resolution and positioning

The optimum performance of analytics are achieved with a holistic approach where all parameters and possible variations are defined. Furthermore, not all processor hardware and analytics software can perform in perfect harmony to produce superior results, as compared to randomly selected equipment that are not optimized or deeply integrated with each other.
In a real world situation, there could be many unknown patterns and complex environments, therefore even “behavioral analytics” performance can be limited as the algorithm demand on the processor can be very taxing resulting in sampling down of the resolution to save computational resources and therefore missing out on recognizing small pixel movement and not triggering a genuine alarm, i.e., of a person breaching a fence at some distance. This makes it outright necessary to plan for additional processor resources before the final takeoff.

The convolution, reliability and usefulness of analytics is rapidly evolving, however what we see in movies and television makes it fuzzy what is realistic and what is fiction. This, by no means translate into unachievable results, as time and again even television shows from as far back as 1960’s like Star Trek are living proof that many ideas and gadgets like the Star Trek “Communicator” or “Tractor Beam” were accurately conceived by Hollywood and either already used in our daily lives today, as is the case with the “Communicator” aka, cell phone or proven to be possible in a test lab, which is the case with the “Tractor Beam”.

What is yet to come in the analytics arena in coming years and decades may surprise many industry speculators, however most of the mature analytics feature available today from reliable suppliers are very useful, practical and beneficial.
Live long and prosper…