Which Video Analytics are Most Important For Your Business?

Blog 6 min read | Oct 5, 2022 | JW Player

Share:

Video analytics is the use of AI and algorithms to perform video surveillance tasks that were once solely the purview of humans. These might include monitoring traffic jams and alerting drivers so they can choose an alternate route, detecting suspicious behavior, facial recognition, analyzing customer flow, etc. 

The technology allows humans to avoid the tedious task of watching video cameras for problems and instead react quickly when a problem does occur.

Which video analytics are most important, of course depends on the nature of your business and your use case. This means it’s important to understand what the types of video analytics are and what they can do for you and your business.

What Types of Video Analytics Are There?

Video analytics fall into three common types:

  1. Fixed algorithm. With a fixed algorithm, developers design an algorithm to look for a specific thing in the video. This might be somebody with their face covered or a broodmare lying down in readiness for foaling. This requires that the behavior be reasonably predictable and that somebody take the time to code the algorithm. Typically, customers have to pay for each specific algorithm. However, for very simple behaviors this can be cheaper than a more comprehensive AI-based solution. It also removes the risk that an algorithm will “learn” an incorrect response and thus react inappropriately. This is still more advanced than simple motion detection, which can only determine that motion has happened, not what it is. It requires extensive metrics to help determine which algorithms you need.

  2. Artificial intelligence learning algorithms. In this case, machine learning is used to allow the algorithm to learn which behaviors to flag. This requires that work be done to train the algorithm. Generally, you purchase an algorithm that has been partially trained for your use case, but have to continue training and monitoring it yourself. The algorithm will learn what behaviors are normal and over time start to flag things which are not normal. For example, an algorithm originally designed to spot kids climbing onto a luggage carousel also learned to spot unusual behavior with bags, preventing theft. However, these algorithms can sometimes learn to flag incorrect things, resulting in false positives, and still need some monitoring.

  3. Facial recognition and similar. These are algorithms that are designed to look at the image and pull out an element, such as a face. Similar code can also be used for license plate recognition, for example to determine if somebody pulling into a parking garage has a permit to park there or to use traffic cameras to find a stolen car.

All of these work in real-time, allowing the system to respond to aberrant behavior immediately. They can also be applied to video footage both to train the algorithm and to analyze an incident after it has happened, such as for training or law enforcement purposes.

What can Video Analytics Do?

Video analytics can do a large number of different things, although it does require some work. This means having the right video analytics software and video management software for your use case.

However, good use cases for video analytics include:

  • Tracking traffic flow through a store to provide an optimum customer experience
  • Personalizing the shopping experience
  • Monitoring for people cutting in line
  • Capturing images of fare dodgers on a transit system
  • Minimizing false alarms for perimeter surveillance, such as animals, which can trigger more primitive motion detection systems
  • Spotting people or vehicles going the wrong way down a one way street or passage.
  • Detecting surveillance tampering such as putting a video on a loop or replacing it with a still image
  • Spotting bottlenecks of vehicular or pedestrian traffic and redirecting people appropriately. For example, if a line gets too long, the system can change signage to direct new people to a different queue.
  • Monitoring and enforcing worker safety protocols.
  • Measuring and controlling occupancy, including enforcing capacity limits or tracking what times of day a building is busiest.
  • Detecting loitering and potentially dispatching human staff to deal with it.
  • Performing object classification and object detection to prevent theft and loss or to alert if an item is not put back where it belongs, which can be particularly important in healthcare.

When we ask which are the most important, it depends on your use case. For example, for a retail store, the most important applications might be to track motion through the store to determine optimum sign and product placement and to monitor check out lines so signage can direct people to different lines and better spread the load. All of these require the appropriate deployment of surveillance cameras to ensure the system has enough coverage as well as the right software for your problem.

What is the Difference Between Video Analytics and Video Surveillance?

A lot of people may argue that all of this can be done with video surveillance. However, video analytics have a huge advantage, and that is that they take the tedious work out of the hands of human beings.

Imagine a security guard monitoring multiple CCTV cameras from a room and directing others to provide physical security for a building. This requires that they continuously pay attention to all of the screens so they can detect aberrant behavior.

In the real world, this is nearly impossible to do. Security guards get distracted, start reading a book because they are bored, fall asleep, etc.

Their attention drifts when nothing is happening and while things can be done to reduce this, such as switching out personnel between monitoring cameras and patrolling, it is inevitable that they will miss something or not react quickly enough.

The use of video analytics is, of course, part of video surveillance. But by allowing the algorithm to monitor the security cameras and then alert human personnel when it sees something, video surveillance systems become much more effective. The number of people needed can also be reduced.

How does Deep Learning Support Video Analytics?

AI algorithms that learn appropriate and inappropriate behavior are far more flexible than fixed algorithms. While there is a risk of them learning the wrong thing, which may then have to be adjusted, intelligent video analytics allows a much more efficient way to develop new use cases.

It allows for a higher degree of automation and a developer does not need to code a specific algorithm for everything you are watching for. This works well for very simple systems, such as detecting when a car drives past and storing the license plate information. It works poorly for, for example, access control and ensuring that people do not sneak in through the exit gate.

Deep learning requires a lot of video data to train the algorithm. This can be done by using video footage from past surveillance efforts or by using footage from a similar use case to “seed” the algorithm and encourage it to “grow” in the right direction.

It’s worth noting that all true decision making is still done by humans. Neural networks have yet to reach the point where they can work unsupervised, and the purpose of the video analytics system is to alert humans when appropriate.

So, what is the most important thing to consider when doing video analytics? The answer, of course, depends on your use case. However, you need to consider whether you have a simple task that can be handled by fixed algorithms, or a more complicated one that requires a deep learning algorithm. This may require that you get the advice of a professional who can help you determine what is most important for you and choose the right analytics platform for your needs.