At a conference on video content analysis (VCA), some speakers queried the hype so far.
But others could point to real-world examples of the products in action. And there we start.<br><br>Barry Keepence, chief technology officer of IndigoVision, reported installations in Essex police stations and airports of video content analysis -VCA for short, also called intelligent video. He did admit to the London event, run in late June by IMS Conferences, that users of some VCA products may be disillusioned by false alarm rates, and simply turn off the product. That said, he did report ‘real-world examples’: “Our experience is that the police love cameras, and he more cameras in the police station the better.” Numbers of cameras in a small and large station may range from three to 35, covering inside and out. Police are, he reported, sensitive to the legal aspects of what they are doing.
Privacy
A prisoner may require some privacy in a cell; but prisoners might quickly learn the part of the cell that an operator could not view. Hence IndigoVision provided video masking; the day to day operator sees the video mask on the private area, while an operator with the right privileges can see the hidden zone. Also installed: a ‘virtual tripwire’ around the car park of a station, because at smaller stations especially there might not be staff able to watch for site intruders. Police also asked for the virtual tripwire in their own server rooms, that is, treating their data centre as a sensitive area. Also popular: audio recorded with video, because video alone might tell a different story than video with audio.
Airport use
Barry Keepence also went through airport uses, mentioning the Doha International Airport (DIA) in Qatar installation by the manufacturer, featured in the March edition. First: counter-flow, in other words people walking in the wrong direction. If someone runs back through a secured zone, even if they have only forgotten a bag, ‘it’s a massive security issue’. Or in arrivals, if someone has forgotten a bag and they return to the baggage claim area. To get good results from VCA in these cases, Barry Keepence spoke of camera angle (in an airport it’s fairly easy to fit a camera high for a good position); and good lighting inside. Also used: what he termed soft barrier monitoring, or the ‘virtual tripwire’, where you want to detect people crossing a line. On airport aprons, you cannot very well have a physical barrier, yet you want to stop baggage carts, even people, if they go where they should not. He admitted that lighting and heavy rain and puddles on the surface cause reflection; but a trial of the tripwire led to a roll-out. He spoke also of congestion detection used in an (unnamed) airport in Europe. Interestingly, this use of VCA does not do behaviour analysis to tell the CCTV operator what the congestion is; merely, the algorithm says that there is a crowd. It is then for the operator to look further: maybe people have gathered around someone who has tripped; or there is a fight. Only two to five per cent of cameras in such a case may run the analytics, he added. He summed up VCA in general: “It has suffered from high and over-promises; but if you cut out the hype, it can work … it has been deployed to bring an operator’s attention to a particular scene or event.” And is there a return on investment? “I think so; if someone goes the wrong way through [airport] security and you can’t track it down and can’t analyse it quickly, you are faced with closing a terminal; that’s very expensive.” If you avoid such a closure once, that makes use of VCA for counter-flow analysis cost-effective, he suggested. At Doha, the video analytics is integrated with the alarm management; that is, CCTV operators treat alarms from VCA like any other door or intruder alarm. What is the future for video analytics? Barry Keepence answered with a blank slide; that is, it’s what people make of it. He foresaw ‘reasonably cautious’ deployment, hoping that the days of hype are over. Installers and end users are getting now a positive experience of VCA, though it is a long way from ‘plug and play’, he added. That is: deploy with care, in a well-defined application. Barry Keepence argued there were two uses of video analysis: live (after an alarm or event), and play-back (to speed a search of data), though the two often share the same analysing algorithms. <br><br>On that note, the next speaker, Paul Swarbrick, Home Office Scientific Development Branch, described the ‘problems’ he has had with VCA. He is the Sussex-based programme adviser – video based detection systems (VBDS), which in the words of the branch’s website are intelligent detection systems which provide automated surveillance through real-time video analysis and event detection. A speaker at the IIPSEC conference in January, he featured in the March issue of Professional Security. Paul Swarbrick said: “We find that very slick sales-people have come along and sold not just CCTV but analysis and behaviour systems that promise the world and don’t deliver anything like the claims o the actual wrapping.” Put another way, the product simply does not do what the smooth talking sales person said. Swarbrick spoke of a sales clip of 30 seconds being in nice weather with no awkward objects like trees around. But a 240 TV lines analogue camera on a lamppost will not perform the same as a 13 megapixel camera used in a controlled test. Performance claims, then, are hard to verify. Yet Paul Swarbrick spoke of police and the military ringing HOSDB after buying such a product that does not match up to the manufacturer’s claims. <br><br>Hence, in November 2006 HOSDB launched the Imagery Library for Intelligent Detection Systems. i-LIDS forms a UK government standard for the development and testing of Video Based Detection Systems (VBDS). Standard video footage has been collected and offered as a library on a disc to manufacturers and researchers, to aid developers of detection systems and algorithms. HOSDB is evaluating commercially available VBDS, starting in July 2007; and Paul Swarbrick reported that the branch is working towards making a mandatory standard. That would mean, if you want to sell your product to government, military and police; get it validated against i-LIDS.<br><br>Alarm reminder<br><br>He ended by reminding the audience of false alarms from VCA alerts, ‘a real problem’,he said. Buyers of products want to have confidence that systems work. He gave the airport example of a perimeter of perhaps 20km. A security response force has limited numbers, as CCTV operator numbers are limited. A user has to have a decent rate of false alarms, otherwise the response force is running around all the time, achieving nothing; hence the product would be switched off. <br><br>Strands<br><br>From the floor, some members of the audience sought to clarify. Chris Norton of perimeter detection product manufacturer Senstar Stellar suggested a difference between nuisance (a tree in the wind) and false alarms; another man observed that installers are slowly changing from electricians to network engineers, and he suggested that the age-old problem, in many industries, is that the manufacturer, supplier and installer are different strands, and the end user does not know who to go to, to fix a problem. The installer (say) may blame the manufacturer for not bringing out a patch yet. Recurring points were the expectations of users; and the gap between the product made in a lab, by scientists, and the operator in the control room who does not have a PhD. Let Barry Keepence have the last word: while most false alarms are from external uses, and after heavy rain, yet VCA products are used on a daily basis, he said. <br><br>Speakers<br><br>Other invited speakers at the two-day event at the Paddington Hilton included Edward Cusack, MD of Guide Security Services; John Williams, head of group physical risk at HSBC Holdings; and Rob Cummerson, marketing manager at Liverpool-based biometrics installer Human Recognition Systems (HRS). At the show, 3VR Security Inc, a San Francisco-based intelligent video management systems product firm, and HRS reported a partnership. 3VR’s CEO and co-founder, Stephen Russell, and Neil Norman, CEO and founder of HRS, gave a presentation on banking at the conference. <br><br>Search data<br><br>HRS Managing Director Neil Norman said “Behaviour of people falls within a variety of ranges for example common, some unusual, some acceptable and some outside acceptable limits. Analysing this behaviour can provide clear indicators about a person’s intention of engaging in disruptive activity.” Speaking to Professional Security, Stephen Russell began with the company’s background; Silicon Valley rather than security – and yes, the company includes some rocket scientists! You can view 3VR clips and broadcasts on the YouTube website. And a 7-7 style search of video data for a suspect’s face he likened to a Google internet search, rather than going through web page after web page yourself. But a user does not have to run the biometrics; he gave the example of cameras at turnstiles. Say the command centre wants to see the faces of people who have passed through the gates in the last hour; hundreds of faces come up. You can click on the suspect’s face, and see who the suspect was standing with. Or, the user can ask cameras to search faces with say a 50 per cent match; that is, you do not require a perfect biometric accuracy. He spoke of diminishing returns by adding computational power, to get more perfect facial recognition. Such video data might, he added, offer the prospect of for example banks checking a ‘watch list’ of suspects against a database, and sending alerts to member organisations.
