Video Analytics and AI Today



Video Analytics and AI Today



Advanced video analytics are available today and offer enormous potential for enhancing or even revolutionising the way you currently perform video monitoring.

Can your current system detect actual adverse events as they happen? Can it track an individual right across your CCTV system from camera to camera? Does your video search allow you to find all images of a person with a blue top? Does your CCTV system just get used for security or does it simultaneously give insight into behaviour of shoppers in your store or whether the layout in your factory is efficient?

This feature aims to introduce you to what is possible today that could increase your incident detection rates and ultimately save you considerable costs on video storage.

Check Your Security limited have been installing enterprise CCTV systems for over 15 years. As an early adopter of IP based solutions, we received a number of awards early on for promoting this technology and today we are still looking ever forward at technologies that improve security operations for our clients. We have seen the nature of video analytics change significantly over this time from simple heuristic-based motion detection and its various applications to the advanced machine learning and artificial intelligence-based algorithms we see today. Technology that now allows us to identify and classify objects such as vehicles, people and even colours.

With the additional intelligence obtained from image data we can integrate with external data sources using Application Program Interfaces (APIs) to obtain additional information that can be stored alongside video evidence. For example, an Automatic Number Plate Recognition (ANPR) system could call an API to return the vehicle make, model and colour. This data could then be searched if you were looking for perhaps a red car or a Ford van now offering two routes to the data after the object recognition has already determined the vehicle type and colour.

This feature introduces a selection of video analytic tools currently available; some may already be available in your VMS as an add-on, a plugin or standalone system. A number of these tools are cloud based and we will discuss the pros and cons of deploying cloud-based Software as a Service (SaaS) tools in your systems.

What is Artificial Intelligence (AI), what is Machine Learning (ML)?
AI and Machine Learning (ML) are two phrases often used interchangeably but they are not the same thing and many products sold as AI may be ML. To the user, the semantics are not as important as the functionality. Check Your Security understand the difference and more importantly we understand their application and how they can help you as a business not only with security but to additionally leverage your cameras to give your intelligence about your customers habits.

Artificial intelligence is a broad concept that can be loosely interpreted as incorporating our idea of human intelligence into a machine. Such machines can recognise and classify objects, react to information from multiple sensors and make decisions based on the algorithms they have been programmed with.

Machine learning is a subset of AI but also one of the techniques for realising AI. Machine learning is a method of training algorithms with a wide range of data such that it can make its own predictions and subsequent decisions based upon its model of the world. Deep learning is the evolution of machine learning whereby a system can determine the features it will use for classification where ML requires these features to be provided.
Applications AI/ML technologies:
Control Room
We cannot expect our control room staff to monitor all cameras all the time, particularly in enterprise installation with hundreds of cameras and sensors. Instead, we can use video analytics to generate event-based alerts enabling our security team to react quickly to any incident. Advanced analytics can be used real time or in searches of archived footage, for example an appearance search for a person with a blue top.

These intelligent analytics process may also be linked to APIs supplying additional information that can be stored in your Video Management System alongside the video evidence.
  • Incident detection
  • Crowds
  • Loitering
  • Behaviour
  • Traffic management (foot or vehicle) atypical behaviour - speed, direction, location
  • Intrusion Management
  • Virtual Tripwires and exclusion zones
  • Access to areas outside working hours
  • Access by people your system does not recognise
  • Automatic tracking across a range of cameras
  • Object Recognition
  • Facial recognition
  • People detection
  • Pre-defined objects in industrial automation
  • Vehicle registration details, vehicle type and colour
  • Worker Safety
  • Wearing of appropriate items of PPE
  • Unsafe behaviours
  • Monitoring of machines and processes
  • Fire, smoke and hazard detection
  • Fluid level and leak detection
  • Behaviour anomalies, injured worker laying down etc.
  • Camera Management
  • Camera tampering
  • Camera image quality
  • Camera recording status and DVR metrics
  • Minimising vibration effects
Retail Analytics
Your investment in physical security rather than being an ongoing cost could be used as a business development opportunity. Beyond simple metrics you can now analyse your customers’ behaviour, see which areas of the store are popular, detect typical routes through your store, find which areas people spend most time, detect returning customers and more.
  • People Counting
  • Face obfuscation / anonymising
  • Visitor profiling – Women, Men Children visiting your store
  • Dwell time
  • Conversion rates
  • Suspicious subject detection and tracking
  • Store and display layout analytics based on traffic patterns and behaviour analysis
  • Logo or brand detection
  • Point of Sale monitoring
  • Smart queue monitoring
  • Heat maps
Integration of video analytics
Advanced video analytics can be applied to your system while protecting your previous investment in cameras and video storage. There are two ways in which video analytics can be applied either at the edge in the camera itself where sophisticated AI tools are chip based and built into the camera firmware or server based that is built into your video management system either on site or in the cloud. We will discuss three integration approaches that will offer considerable improvements in your security management systems:

Integration of tools in your existing Video Management System (VMS) software

Replacing your existing VMS

Cloud based add-ons
Maintaining your existing system
In many cases this may be the most cost-effective route to the integration of advanced video analytics. Your team may already have several years’ experience of your current system and you may have a considerable enterprise-wide investment in this kit. You also do not want the scale of change that means re-training all of your operators and teams that maintain your existing infrastructure. Most enterprise VMS have an integration capability whereby plugins can be used to add additional functionality to the users existing screen-based console. These plugins will have access to your live and archived video to overlay an AI capability on your existing hardware. AI and ML technologies can be processor intensive so typically these add-ons are implemented in two ways; A dedicated analytics server with AI specific processor hardware that runs the analytics tools without impacting your VMS servers or using a cloud-based integration whereby the massive compute power available on the cloud is used for your analytics processing. The onsite solution clearly allows you to maintain your CCTV as a closed system and avoids using up network bandwidth streaming images offsite.
Replacing your existing VMS
The most flexible solution may be replacing your current system particularly if your existing system is either due a lifecycle refresh or is simply not performing how you would like. Like any new investment, you need to be sure you know what you need rather than having a system sold to you. Do you need or wish to maintain your existing cameras? Can you replace a small number of cameras at key locations with advanced edge analytics model? Should you replace everything because of the additional benefits the new system may offer? Should it be on site or cloud based? Is the system open to future integration and changes? Can I use the system for more than just security? Are you locked in to a single vendor?

Our experience at Check Your Security can help you decide from the ever-increasing choices, which products are best for your unique situation. We will also assist in the operation and maintenance. Some of the most important choices today are around flexibility and total cost of ownership over the projected lifetime of your system.
Cloud-based add-ons
Significant processing power is required for many AI/ML functions and dependent on your situation you may find that onsite solutions are expensive to maintain or scale. Cloud based products on the other hand can scale seamlessly and in many instances be upgraded with no disruption to the end user. There are pros and cons with any of these solutions, for cloud-based add-ons you will be required to have an off-site connection to the vendor. Whilst in transit security and encryption are robust, it may not sit with an organisation’s security policy although cloud computing is more often than not more secure than onsite solutions. Numerous cloud-based solutions are already being utilised by MOD where security is paramount and not negotiable. Being service based cloud solutions can be flexible so for example if you only require the analytics tools infrequently or for a campaign does your vendor offer short term licensing options and similarly you are free to switch if you find a better service and can cope with the change management for your users.
CCTV system efficiency gains and cost savings
There are still system operators out there that have not fully moved on from their days of running analogue CCTV systems. They are most likely using digital storage but are streaming their cameras at 30FPS 24 hours a day recording empty corridors and rooms at great expense in terms of storage and network bandwidth. Simple pixel-based motion detection for many years has allowed system operators to run cameras at a low frame rate until motion is detected, the frame rate then increases and full speed video is stored together with a pre-event buffer of seconds before the event to ensure nothing is missed. Today this simple motion detection can be set to trigger only for people or other specified objects avoiding false triggers, real events not just changing light conditions or clouds moving as with simple motion detection. Does your control room have a video wall that operators are expected to monitor 24 hours a day? Given that around 20 minutes is the average time an operator can stay focused and alert or an interesting non-event may be more distracting than a real event is your video wall just a vanity project? Correctly configured AI based alerting can minimise human error significantly and make considerable efficiency savings. For unmonitored system it will offer small businesses a virtual security guard.

Event driven, uncluttered control rooms are common in many business sectors today, but many legacy CCTV operators are yet to realise the benefits. Poorly configured event driven solutions using basic motion detection have turned many users off forcing them to revert to what they are comfortable with, sitting an operator in front of 20+ screens. Events are uncommon in a typical day so your system should reflect this, you may get the occasional false alarm, but you will capture considerably more real events than before, giving your security team the tools they need to do their job well.


UK GDPR and video surveillance

It is already a requirement for operators of CCTV systems to understand that data collected through video surveillance is considered as personal data and the guidelines for processing of personal data that apply.

Facial recognition is a particular case that is recognised as sensitive data and as such specific consent is required to use this technology. Detecting a face using AI is not the same as using a facial recognition system to uniquely identify an individual. According to the Information Commissioners Office (ICO) “You need to identify an article 9 UK GDPR condition, if you are actively processing special category data, such as biometric data (for example when using facial recognition systems to uniquely identify individuals).”

In terms of integrating surveillance technologies, it is important that you can verify information collected is: Use is necessary and proportionate throughout the lifecycle of processing.

System integration may also have implications relating to profiling individuals that help your system learn and make decisions about them.

In all instances you should perform a Data Privacy Impact Assessment (DPIA) that will help inform your lawful application of these technologies. More information and some useful checklists can be found here.

Kevin King Head of Technical Services Check Your Security Ltd