Video Analytics: Separating Fact & Fantasy
Something strange is going on in video surveillance – a gap is opening up between what’s loosely desired by end users and what can be achieved by video analytics in the real world, with some consultants writing specifications that seem to incorporate too many analytics with scant regard for how such solutions might be delivered.
A growing number of quality security integrators say they are not prepared to attempt to meet ‘specifications’ that are not grounded in clear operational requirements and functional possibilities. Their response is to withdraw from such jobs and, in many cases, they are beginning to lose faith in analytics as a technology altogether.
On one hand, it’s not surprising that end users in technically advanced societies, particularly those with a hunger for tech’s bleeding edge like ANZ, should tick boxes for any number of analytics, the idea being that a CCTV system has the potential to act as first alert for…well…almost anything conceivable, if you can design it, if they can pay for it.
There’s no question that video analytics, further empowered by deep learning, offer more and more possibility. Deep learning works by parsing monstrous amounts of data to tune circuits until they are expert at identifying multifarious sums of data inputs. The more data they process, the more expert they become.
Deep learning may offer users surveillance solutions that can recognise faces, gender, gait, moods and events that breach its vast, collective experience – many people running, gunshots, chemical signatures exceeding background thresholds, vehicles where they should not be, outbreaks of fire, groups of people in conflict, traffic accidents, medical emergencies, elevated temperatures, abusive signs or word, or any variable deliverable by any conceivable sensor input.
But taking proprietary systems with set parameters out of the equation, how are such analytics to be integrated by our technicians? How will they be designed and commissioned, who is prepared to pay for the huge servers required to drive them? And what is the real operational benefit of customised analytics – where is the return on investment?
Making things harder is that video analytics is vertical specific. For retailers, it’s data about shoppers that allows them to understand what’s selling and where. For big operations like airports, it’s about efficient movement, minimisation of pedestrian and vehicle traffic jams and avoiding congregation points on the vulnerable public side of the terminal.
For defence installations, analytics is about policing boundaries, recognising intrusion day and night with absolute surety, and allowing the largest area to be policed by a very small, highly mobile, security team. For city managers analytics is something else again – it’s about vehicle flow, managing pedestrians and ensuring large gatherings are policed. And for a stadium or a casino, analytics is about recognising a known offender and actioning fast response to bad behaviour.
Then there are applications that defy norms. A commercial site may want LPR, face recognition, boundary policing, and real time event detection because such things are possible in one-out vertical applications. But trying to wrangle multiple analytics applications and to deliver their actionable events to the operators of one management system in a coherent way is challenging in the extreme.
Something else that’s tricky is that analytics drive through VMS but VMS makers often aren’t the analytics people. Even if there are analytics integrated into their solutions, VMS makers will partner with specialists in a branch of analytics. This moves the relationship a step away and increases costs and project complexity, as well as introducing trust issues.
Trying to understand how an end user’s opaque desires to leverage analytics might be delivered via an external software house in a way that allows possibilities to be synthesised and delivered affordably is what’s at the heart of this issue. Video analytics solutions for major systems, particularly those in public spaces or high traffic areas, are simply not plug and play.
There are positive takeaways from the current situation. The first is that end users are interested. The second is that consultants are prepared to give end users what they are asking for. But it gets stickier. Caught between rubber and road – forced to come up with solutions delivering improbable functionalities using uncertain delivery mechanisms, all while being shivved by underquoters unlikely to be held to account by end users who don’t know better – serious security integrators are increasingly frustrated with the entire process.
In the future, electronic security sensors, including CCTV cameras, will have more powerful processors, giving them greater capacity to inform users using more communication paths. These sensors will feed breaches of threshold events via edge devices into a management system, locally or via cloud, from which managers can derive unprecedented situational awareness.
But how will they do this? Who will design and engineer these systems, who will integrate them and is anyone prepared to pay for the information they can offer? These questions need concrete answers. If we don’t want end users being led up the garden path, it’s high time consultants, manufacturers and manufacturers’ partners sat down with security integrators and started talking straight about what is and isn’t possible with video analytics.