printer friendly version

Automated analytics with iSentry

There are many respected video analytics systems on the market for surveillance users to choose from, some developed right here in South Africa. The critical component in all of these systems is the ability to analyse video footage and initiate an action or response when a suspicious activity is noted.

One of the systems is iSentry from ISDS, although it has its roots in Australia where it was developed by Sentient Vision Systems. What makes iSentry different from other analytical systems is that instead of pre-programmed algorithms that determine if something in front of a camera is of interest, iSentry is designed to be self-learning.

As noted, iSentry originated in Australia before being brought to South African ISDS in 2008 and jointly developed thereafter. The original idea for the system was to provide visual input to machines in order to give them a form of awareness without requiring human input. The company then moved into other areas where the same technology could be used effectively, namely surveillance.

One of its first customers was the Australian Roads Agency, where the organisation wanted to be informed of road incidents in real time. This was the beginning of the solution now known as iSentry.

Callum Wilson, MD of ISDS explains that in the road agency scenario, the organisation needed to catch any anomalies in the traffic, whether an automobile accident or something involving a pedestrian. This is a critical environment and missing something is never acceptable. It also means that the management system needs to be able to detect expected problems as well as those that have not been specifically programmed.

iSentry beat out its competitors in this scenario because its analytics are not rules-based. This means it can teach itself what qualifies as normal, and highlight any anomalies irrespective of whether they have been seen before. In the Australian Roads Agency project, the system was the only one that detected that an aeroplane making an emergency landing on the highway was an anomaly. The other systems accepted the plane landing because it was landing in the direction traffic was supposed to be going and the pilot was able to avoid causing an accident. iSentry was able to ascertain that this large object coming out of the sky, even though it was obeying the rules, was out of place.

Wilson explains that the system does not grade events, but rather learns what is normal and highlights abnormalities. Parameters are set during installation, thereafter the system builds its own intelligent decision making processes. Currently, iSentry handles behavioural analysis, object detection, moving target detection at land and sea, and ground change detection, among others. Wilson says it is very effective in wide outdoor environments, even in adverse weather conditions.

Independent learning

When using surveillance systems to monitor fluid and busy high-risk environments, using a rule-based system is difficult. If something happens that falls out of the set parameters, no alert is given when an operator should receive one. iSentry avoids this by learning what is normal and then alerting operators when anything falls outside of this norm regardless of whether the event is in fact a critical scenario or not.

Wilson says there is no human interference in the learning process and no prior learning is required. The system really analyses the situation itself and applies hierarchical artificial intelligence to determine what is normal and what’s not. It analyses the scene from a pixel level and decides what belongs there, what moves, what doesn’t move and in which way they move. It then marks anything outside of the paradigm it creates as unusual.

As an example, Wilson says that setting the system up to monitor a busy road and leaving it for 48 hours will see iSentry learn enough to cut out 95% of the traffic on the road.

The key, according to Wilson, is to never miss an alert, but to make the whole surveillance process more manageable. Instead of having operators watching a host of screens, the system allows them to ignore 95% of the footage and focus on the 5% it determines is unusual. And as operators dismiss some of the 5%, the system continues learning and reducing the number of alerts it provides.

It’s also worth noting that the alerts are delivered on live video, not recordings. However, when something of interest is raised, the system allows operators to drill down into historic footage if required.

Camera choices

As with all management platforms today, iSentry branched out from analogue cameras into the IP world and initially connected to a variety of cameras via the software interfaces provided by the manufacturers. This, however, proved less than optimal as code had to be developed for every camera supported, and then updated when the firmware on the camera was updated – a never-ending task.

The Australian developers then decided to take the raw RSTP (Real Time Streaming Protocol) video stream and analyse that. After some work, the software works effectively and iSentry is now effectively hardware agnostic. For remote monitoring applications, ISDS has also developed its own compression algorithm to limit bandwidth usage. The proprietary streaming protocol also adjusts to the available bandwidth to ensure the images get through to the control room.

The hardware agnosticism also applies to the systems iSentry is run on. It can make use of commercially available IT platforms if required. ISDS, however, also provides a full solution to support its clients and resellers.

The application’s interface is designed to be easy to learn and use, while being intuitive. When an alert is raised, it remains on the screen until the operator dismisses it, ensuring someone takes action and is accountable. It also keeps a timeline of the events of the day (or however long the company requires), allowing supervisors to quickly move through historical footage and alerts.

The different situations operators may face can also be organised and handled by means of standard operating procedures (SOPs), which will ensure operators take the correct actions every time.

Making the case for CSS Tactical

CSS is a security service provider that started operations about eight years ago in the Illovo area. The suburb was using traditional armed response services and finding it ineffective. CSS decided to take a different approach to guarding the suburb with tactical vehicles and better trained personnel.

Even with the improved equipment and people, guards can’t be everywhere all the time. The company therefore decided to combine people and technology and install cameras (with the residents’ approval) to enable it to conduct live surveillance 24x7. CSS also realised that having people staring at a bunch of screens would not be an effective solution and needed a software solution that would filter out the irrelevant data and alert its operators to anomalies.

CSS chose iSentry and now sees about 97% of the video surveillance footage ignored by the software, with anything unusual, such as a prowling car late at night, brought to the attention of operators who can then dispatch a unit to investigate. The fact that iSentry keeps an audit trail of everything the operators do, provides CSS with an additional ‘guard’ that watches the operators to ensure they are not compromised.

CSS has since taken this approach into other areas, such as Dunkeld, where it has seen contact crime decrease from one per week at the start of the contract to one every four months, proving the effectiveness of its approach. Using iSentry provides operators with increased situational awareness, which allows them to make optimal use of their human resources on the ground and see to it that the armed response teams arrive before the criminals can act.

More information is available on YouTube at: www.youtube.com/watch?v=iu23WQra8to and www.youtube.com/watch?v=yjnQmFP9TGQ

Share this article:

Further reading: