Analytics as an event trigger
Today’s surveillance market has an embarrassment of riches when it comes to the creation of smart solutions. Despite this, implementing systems that solely deliver security-based performance still makes up the vast percentage of applications. Given the ability to filter events, subsequently enabling the triggering of other systems and automations, are installers and integrators missing a trick when it comes to video analytics? Benchmark considers the role of IVA in site management.
For many years, the promise of intelligent video analytics and all it might deliver in terms of customer benefits has been something of a driver for video surveillance manufacturers when investing in research and development. Despite video analytics having been available for over a decade, and regardless of the many advances that have been made in that time, intelligent video analytics (IVA) is still perceived as something of a security-focused technology.
Indeed, even just a few years ago it would have been seen as foolhardy to suggest that the technology had a future performing basic site management tasks. The software was complex, hungry with regard to required processing power and costly. It could be argued that IVA actually wasn’t too expensive for site management roles; it was more a case that many in the security sector did not understand the value end users would place on the automation of certain business tasks!
As manufacturers worked tirelessly to improve the accuracy of video analytics and to add filters and discriminations that enhanced its stability, it almost went unnoticed that some of the problems analytics presented for the security sector could potentially be benefits for automation and management tasks.
The reality is that video analytics works by detecting exceptions. In other words, video analytics is looking for behaviour that is not expected. For example, if a camera is capturing video of a fence line, the user does not want to know when ‘normal’ behaviour is occurring. If people are walking, running or stopping, it doesn’t matter as long as they remain on the public side of the fence. What the user wants from IVA, in a security sense, is to know when someone crosses the fence line.
On the face of things, that sounds simple enough. If a sterile zone exists, the situation is easier to deal with as the analytics simply need to detect a person on the protected side of the fence. However, if a sterile zone is not in place, and there are authorised persons on the secure side of the fence too, the analytics must detect the action of a person crossing from the public side to the secure side.
The issue with this scenario is understanding how people will cross the fence line. They might climb it, cut through it or burrow under it. In a very high risk site, they might even ‘drop’ into it. All eventualities must be covered. Alternatively, the might use a legitimate entry point such as a gate, and then deviate from an expected route.
Cover such as buildings, bushes and trees or vehicles might be used to conceal the entry. Additionally, those intruding into the site will conciously take all measures to make their actions seem to be normal.
As such, the reality is that in a security application, even the most basic IVA task can quickly become complex. Security is inevitably protecting against the unexpected, and as such video analytics are required to quickly and accurately detect the unknown.
Known and unknown criteria
In order for an IVA engine to accurately and consistently detect threats, it has to know what it is looking for. As we’ve established, in many cases the ‘known’ criteria will be ‘normal’ behaviour. In order to detect the unexpected, the IVA needs to be proficient at detecting exceptions to this.
This is a more difficult task, and in order to address it many manufacturers will strip down analytics rules to detect one behaviour per rule. This simplifies things somewhat, as the range of exceptions can be dramatically reduced. However, it can also mean that the usefulness of some IVA implementations are equally reduced.
The outcome can be that in certain applications a number of different analytics rules are needed to deliver the required level of protection. Combining multiple rules is an acceptable approach with some IVA technologies; with others it can be a problem waiting to happen. Often field trials are the only way to find out exactly which rules complement each other and which don’t. For installers and integrators, it can be a slow and frustrating task.
IVA has restrictions, and understanding the restrictions can be the difference between a successful installation and a system which does not meet end users’ requirements. In many cases, the correct configuration of IVA can be time-consuming, and where challenges are faced due to the use of multiple rules the resources required to ensure optimum performance may be significant.
Of course, these restrictions should not be seen as a negative when considering the deployment of intelligent video analytics. They do need to be thought through, but the reality remains that IVA is a powerful tool for security-biased video surveillance. Also, the issues are becoming increasingly less significant as new technologies emerge.
The drive to advanced processing and the wider use of GPUs has made deep learning and artifical intelligence a reality. The benefits for IVA include a degree of self-learning, and the ability for users to interact with the analytics engine to ensure that the needs of specific sites are met. As more data is processed, so the analytics engine is increasingly able to spot violations and exceptions, and to relate these to other data. In terms of security-biased analytics, the drive forwards is only just starting.
However, whilst security currently demands that largely unknown criteria form the basis of detections, automation and site management predominantly use known criteria. For intelligent video analytics, this makes things so much simpler! For installers and integrators, the creation of effective rules is also a more straightforward task.
For example, consider a conveyor belt in a factory. As part of a process control system, the user wants to know if that belt stops or is moving in the wrong direction. They might also wish to know if an item is handing off the edge of the belt, or if an item appears or disappears.
Because the belt is in a controlled environment, lighting isn’t a great issue, and the likelihood of nuisance alarms is minimal if not non-existant (unless a system fault occurs). While an animal setting off an IVA rule at a perimeter would be a nuisance alarm, an animal on the loose in a production area would be a valid concern!
Because the environment can be controlled and the behaviours being detected can be clearly defined, the analytics rules are subsequently simpler to configure. Additionally, because the cause of an IVA alert is known, this means that cameras can be positioned specifically with that alert in mind.
Another benefit for the installer or integrator is that IVA settings can potentially be replicated across a number of zones on the conveyor belt, as well as to additional conveyor belts in the factory.
Monitoring flow on a conveyor belt is one simple use for IVA in a business management scenario. However, for the company using the technology, it protects their most significant activity, the core activity that earns them their revenue: manufacturing. This means that the end user sees significant value in the use of the technology.
Logistics companies could use IVA for parcel and pallet tracking, stock management and vehicular traffic control. Depots could use it for health and safety compliance when moving items around the site, access control via ANPR for fleet vehicles and management for loading and unloading. Transportation sites could implement safety systems such as detecting persons too close to platform edges, as well as vehicle management.
Effectively, if a member of the management team can visually identify or verify a site status condition, this can be automated via the use of IVA. As such, the deployment of IVA can be used to identify status conditions relevant to the management of the site, and then trigger actions or events to enable a more efficient approach to dealing with the situation.
One important thing to bear in mind when considering the deployment of IVA as a trigger in business applications is that you are not always looking to create an alarm. Think more in terms of a cause (the initial detected event) creating an effect (the subsequent action). This might take the form of a notification to specified staff, or a more general broadcast of a message or alert, but equally could be the switching of another system that subsequently delivers the required action.
For example, if an overheight vehicle is detected entering a warehouse site, the IVA activation could then trigger specified barriers to ensure that it can only travel to a loading bay with sufficient height clearance. Such an action simply manages the situation, delivering everyday efficiencies to the business.
Installers and integrators must ensure they have a good understanding of how the user wishes the IVA to perform, as on many sites one type of event could require multiple actions. There might be a need to employ scheduling, so that events are treated differently during site open hours and closed hours. The user might want different actions to be initiated dependent upon staffing levels, or even which specific staff members are on site.
A good example of how an understanding of the user’s needs can win contracts comes via one of the Benchmark test team. His company was pitching for an installation at a factory. On the site was a large department applying coatings which required a number of energy-hungry fans. It was suggested that these could be switched off if no one was in the area.
The user didn’t seem to care about saving on energy costs, because they had a concern that if someone was working without the fans there might be a significant health and safety risk. They preferred that staff manually switched the fans off as they could then double-check the area was vacant. If this task was forgotten, it was a small price to pay.
Picking up on the need for health and safety compliance, the installer then suggested that the IVA could automatically switch on fans if someone was sensed in the coatings department, thus eliminating the risk of them not being switched on at the start of a shift. This idea was accepted without hesitation.
Many installers and integrators (and many end users too) still associate IVA with high costs. While it is true that some IVA implementations are costly, both in terms of capital investment and configuration time, these are generally advanced systems for security applications, where they are detecting the unknown. Because triggering systems for automation in business use-cases tends to use simpler known criteria, it therefore holds true that simpler IVA can be deployed with great effectiveness.
The advances in processing power in recent years mean that today’s edge devices have significantly more power than is needed for just streaming video. It’s a far cry from the days when HD1080p video was a challenge and many manufacturers had to either limit frame rate or bit rate! Open platform cameras and codecs that can run multiple third-party applications are common, as is the inclusion of licence-free video analytics on many devices.
This ensures that IVA implementations can remain cost-effective. Additionally, because of this there is less need to try and use cameras for dual purpose. Often many IVA business use-cases require specific camera set-up in terms of the device location. The addition of cost-effective devices with integral IVA makes sense and can easily be installed on a separate network.
Where devices are being used to simple switching tasks, it may be sufficient to use camera outputs. However, where there is a need to deliver highly flexible solutions, a VMS with event handling capabilities can be deployed. These are available at a range of price-points, including some options which are free-of-charge! Benchmark tested such VMS products here.
It is too easy to limit the application of video analytics to security-specific applications, but increasingly end users are looking for added value that has a return on investment. Enhancing business efficiencies ticks many boxes for them. For installers and integrators, advances in technology in recent years makes the delivery of such solutions simple and cost-effective.
Increasingly, the rise of deep learning and machine intelligence makes the delivery of high-level analytics a reality, and with that comes great interest from the IT sector. Companies in that space are only starting to learn about the potential on offer; installers and integrators already understand it.
Unless the industry takes the step forwards now, it could see the potential for IVA in business applications quickly swallowed up!