We’re all excited by analytics

 

We’re all excited by analytics

 

Face it; we’re delving into the realms of science fiction here: computers that understand what They’re looking at. Although we’re not quite there yet, there is still a place for this remarkable technology and as the software improves and computers get more powerful, it will only improve.

 

VIDEO ANALYTICS is one of the most exciting fields of development in the CCTV sector.

Admittedly, none of us wants to appear too enthusiastic, or look foolish in our expectations – not like the police inspector who recently  asked one company if their software could spot concealed weapons in video footage (to which came the common sense reply ‘well, can you spot concealed weapons?’ – obvious, we’d never have asked that one…)

 Certainly anyone who has followed developments over the last three or four years has taken on board the ‘expectation management’ message from developers. And yet - analytics have come such a long way in such a short time that much more impressive things are surely waiting just over the horizon, aren’t they?

 

Eye on the horizon

But what are the short term development prospects for analytics systems, and is it worth buying them now rather than waiting a couple more years? The answer to the second question is undoubtedly yes, analytics can be a useful tool right now for a growing number of applications.

Our expectations may have been ‘managed’ but it’s clear that the best systems already do some jobs very effectively: spotting intruders in ‘sterile’ zones, counting people or vehicles, identifying when something moves that shouldn’t move (or vice versa).

The viability of analytics techniques – the fact that they now offer practical solutions to real problems - is being driven forward not just by increases in computing power or the refinement of algorithms, but also by the fact that the latest cameras and surveillance equipment produce higher quality images for the software to work on.

Further, companies such as Ganz Solutions are putting forward some good, practical ways that analytics can be used. The latest edge based Ganz VCA solutions can help save money, or bring in more revenue, and the company hopes that this will appeal to cash-strapped local authorities and other major system users.

One of the most knowledgeable practitioners working in the analytics field is Geoff Thiel of VCA Technology whose software is among the front runners and is used by a number of major CCTV equipment manufacturers as the basis for their analytics solutions.

Geoff sums up the current level of development with a neat image: “If the goal is for your system to be capable of analysing video as well as a human being, then we’re at about mouse level. What I mean by that is that if a person wants to fool a mouse, all they have to do is sit very still. If the person doesn’t move then the mouse won’t ‘see’ them. That’s about the level of sophistication that analytics have reached.

” The gulf that separates mouse vision from human vision is a massive one – to get from one side to the other entails a step-change improvement from just ‘seeing’ to actually acquiring some sort of understanding

 “We might be able to get to that level, if you give me a computer that’s ten thousand times faster and then build in all the knowledge that I’ve been learning from the moment I was born. But then if we could do that we’d have robots doing our jobs as well. Those developments would have much more massive implications for automation and our society than just how well analytics works!”

 

Fantasy & reality

It’s useful to keep this perspective in mind when thinking about why analytics systems find apparently easy tasks impossible to deal with (and why those scenarios in TV dramas really are fantasy).

If you place a toy car on your windowsill you can make it appear to be the same size and shape as a real car parked outside on the road. To us the difference is obvious and we can see at a glance that the two are totally different things.

But our ability to make this apparently simple distinction is based on very subtle visual clues such as reflections in the glass, shadows around the objects, the different quality of light inside and out, and colour subtly reflected from one object onto another. We are not always consciously aware of all these ‘cues’ but we still interpret them in an instant. Crucially, we also use context: we know that there are such things as toy cars, and real cars and we know lots of small ways in which they look different.

 Geoff gives an example of an apparently simple analytics task that proved much more challenging than expected, and almost cost one system supplier its incumbent position with a large retail chain-store customer. “We were contacted by the company which was very worried that their current analytics weren’t working at certain times of day in certain stores. The system was supposed to count people coming into the stores but it was getting massively exaggerated footfall-counts in the mornings or evenings as the rising or setting sun streamed in through the doors. The problem was being caused by movement of the automatic doors causing sunlight reflections and shadows to move across the floor in the detection area. Worse still, moving the camera and detection area was not possible because during the day the shadows and reflections moved around with the sun angle making the whole of the shop front a no-go area for the analytics.

” VCA’s ‘Counting Line’ algorithm, which is one of its three key analytics solutions that it offers to the market, solved the customer’s problem and is particularly in demand for retail applications.

“For these counting tasks you need a very tolerant system that can deal with difficult lighting conditions. Another problem is high reflectivity – light thrown up from shiny floors, for example. We developed an algorithm that is very good at counting people.

” Other applications include in the transport sector. In a recent test, the VCA Technology algorithm achieved 95% accuracy counting people through an airport doorway, and neither was it fooled by passengers wheeling suitcases behind them, or by people coming through in clusters.

 But there are limits. Geoff describes another challenge, and this one was too difficult to overcome by just using analytics, the camera location had to be changed as well.

“We came across a situation recently that completely defeated the analytics. The site in question had an eight foot high security gate that slid across a double roadway. The user wanted an analytics system to monitor how often the guards came out of the gatehouse, to verify that it was at least once every half hour.

” This sounds simple, but it wasn’t. Every time the gate opened the system was fooled for about a minute. This is because it could no longer rely on the basic analytics technique of separating the foreground

from the background – the change in gate position looked like a massive change to the background.

 “If you know how analytics works, you won’t be surprised that we couldn’t solve the problem. In the end the only answer was to move the camera to a downward view so that the gate was not in the detection zone.”

 

What now?

So with these caveats in mind, how can users take advantage of analytics now?

They can choose between installing analytics at the centre – with software running on PCs or DVRs in the control room - or at the ‘edge’, with the latest IP cameras featuring analytics capability too. Both routes have their advantages. The choice will probably be dictated by whether you’re building a new system, or just want to add analytics functions to a few locations to achieve specific goals or perhaps experiment with how well it works, and how it well it improves operations.

 For example, all Mitsubishi’s new IP cameras come equipped with basic video analytics, which includes an advanced tracking algorithm promising low false alarm rates, real time data tracking and detection zones.

The company also offers an advanced package that can be purchased as an optional upgrade that comes equipped with behavioural detection, 3D behaviour, counting functions and meta data.

 “Our video analytics offers a range of innovative features,” says Dean Kemp, Sales Manager. “For example, it can identify different objects (from people to animals) and monitor them accordingly. It alerts users of suspicious behaviour by detecting different movements including direction, stopping, loitering, entering, exiting, appearing and disappearing motions. The 3D behaviour feature evaluates images and puts them into perspective, improving the accuracy of the surveillance which is particularly useful when reducing false alarms.

“The advanced tracking algorithm enables on screen monitoring without the need of a security guard. Overall running costs and time can be reduced dramatically, as a 24/7 security guard is not needed.

” Other features promoted by Mitsubishi include an easy to- use web browser interface, early warning alarms, real-time data tracking and people/ vehicle counting facilities. Mitsubishi’s video analytics have passed the i-LIDS test and have also received several certificate tests such as CE and FCC – all useful standards that can serve as a starting point for anyone assessing the various market options. (Developed by the Home Office Scientific Development Branch, the I-LIDS standard assesses how well analytics systems perform using CCTV video footage of four scenarios: abandoned baggage detection; parked vehicle detection; sterile zone monitoring; and doorway surveillance).

 

Black wall monitoring

Ganz analytics achieved certification for Secondary Detection in the i-LIDS sterile zone monitoring test.

The sort of practical application being advocated by the company includes ‘black wall monitoring’, an approach which is designed to reduce the monitoring workload and, in some situations, allow control rooms to scale back on staffing.

While, as we have seen, software cannot come anywhere near the ability of a human beings to interpret or even understand what is shown on screen, it can be used to filter out images which do not meet some basic criteria.

 For example, says Santosh Muniswamappa, of Ganz Solutions, when there is no movement on the scene in a defined area, the image does not need to be displayed on the monitor wall.

 Only when movement is detected (or when another criteria is fulfilled, such as an item being left where it shouldn’t be) is the image fed through to the monitors. Using this technique there is an increased chance that priority images will attract the attention of the operators.

 “This smart way of using surveillance eases mental strain on CCTV operators, creates more proactive security systems and improves the overall operational efficiency of the system.

”Ganz offers a comprehensive range of detection filters which provide the means to discriminate between different object behaviours, depending on the scene being watched. This allows system designers to choose how the solution works.

“Easy to set up, GANZ VCA automatically adjusts to the conditions it encounters, meaning that it is ideal for both experienced and untrained personnel to use.

” Users can choose between a useful list of features: intrusion detection; vehicle monitoring; abandoned object detection; people counting; loitering detection; tamper detection; object removal; and shape based detection.

Looking forward, Ganz also raises the possibility of using analytics to upgrade CCTV cameras to allow them to be used as speed cameras, either by switching to smart IP cameras or simply by adding an IP video encoder to existing legacy units. Analytics would allow councils to get much better value from much of their public area CCTV infrastructure, he suggests.

“This kind of innovative approach would give the councils a dual edge, potentially allowing them to use their numerous CCTV camer-as for speed violation detection as well as surveillance. There would be no requirement of film or local media as these images can be transmitted remotely. This technology can be extended to not just speed detection, but behaviour monitoring (loitering), left object detection, parking violations, one way breaches, red light violations, etc. This will allow councils to get a better return on investment from legacy infrastructure already in place.”

 

Camera choice

Ganz also stresses the importance of linking analytics to the right hardware, because it can do much more with the right cameras.

 For example, in one project for a nuclear power plant the company has transformed the effectiveness of analytics by replacing day/night cameras with thermals.

“In this case analytics are being used to detect intruders at key perimeter locations, but noise from the day/night cameras as light levels fell was generating huge numbers of false alarms. By using thermals instead we have solved the problem. The intruder detection analytics work perfectly.

” In another, industrial application, a thermal camera is being used to monitor hot smoke as metal is smelted; an analytic algorithm triggers additional lighting at the site whenever the density of smoke crosses a specified threshold. The enhanced light allows conventional CCTV monitoring of the smelting to continue.

 It’s also worth paying attention to the software interface and processing because, with good systems, more can be made of the analytics features.

 For example, Ganz smart NVRs allow playback video to be treated with analytics as well as live video. As a practical example, this means that a search can be made of recorded footage with analytic functions retrospectively applied to the recordings even though they were not originally used – to identify when a particular object moved, for example a car being stolen from a car park.

Analytics are also being used to create more robust, multiple-alarm functions with the Ganz VSOIP suite. With this approach, several detectors or types of detector can be linked with video analytics; an alarm is only triggered when a predefined number of detectors is activated, resulting in a more robust system.

 

Analytics as standard

Looking at what’s on the market at the moment – and most of the big manufacturers are now offering some analytics in some form – Geoff Thiel says that buyers have some fairly simple choices to make.

The leading developers are pretty much neck and neck, he points out, because they are all relying on the same computing power and facing the same basic constraints.

The surest approach is for planners to test systems for themselves, particularly if they are considering a major investment or worried about making a poor choice.

“Our algorithms are regularly tested against other major developers - such as IoImage and Bosch - and we normally come out ahead, so we are confident that our system is as good as the best.

” Certainly be sceptical about over-optimistic claims, he cautions – but on the other hand don’t be too sceptical or you may overlook the very real benefits that analytics systems can deliver today. And you are most likely to reap benefits in cost savings and efficiencies if you have a clear specific objective in mind that matches the already-established strengths of this developing technology.

 

 Mitsubishi cameras feature on-board intelligent video capabilities

 

AS TECHNOLOGY continues to develop so does the need for an advanced security system that not only provides reliable security surveillance but also benefits from a range of value added functions; that’s why Mitsubishi Electric, a world leader in the innovative design and manufacture of enhanced security products, has launched a new range of fully functional IP solutions.

 Built to the highest standard Mitsubishi’s new IP cameras are all network ready enabling on-screen security monitoring via the internet, regardless of distance and location. Ideal for multi site locations, our range of indoor and outdoor cameras are MPEG-4 and H.264 compatible, feature full frame rates and quality settings up to D1 and comes equipped with the very latest video analytics (standard analytics built in – advanced video analytics package available as an optional upgrade).

 Over recent years, video analytics has become increasingly popular within the security market. With so many CCTV cameras within today’s society, on screen security monitoring can be timely, costly and inefficient. Mitsubishi’s video analytics package has been designed to eliminate these issues by using dedicated software to monitor real time CCTV surveillance and proactively respond to different characteristics and motions. With a range of innovative features, video analytics is able to detect and identify different objects such as people, vehicles, trees and animals and track these objects accordingly. If the software monitors a movement, behaviour or action that appears suspicious it is able to warn the user immediately. At the same time it is also programmed to ignore irrelevant information like a tree blowing in the wind or a bird flying, reducing false alarm rates dramatically.

The 3D behaviour feature puts images into perspective by evaluating the size, distance and speed of movements, improving overall detection, accuracy and reliability of the analysis, whilst the advanced tracking algorithm enables on screen monitoring without the need of a security guard.

 Easy to use, Mitsubishi’s video analytics comes equipped with a user friendly web browser interface. Furthermore the software can also be used for market research, with a people and vehicle counting function which uses specified zones to calculate the busyness of an area.

Overall Mitsubishi’s video analytics is more reliable than a security guard; it eliminates human error and reduces overall running costs. For further information please call Mitsubishi direct on 01707-278684.