29 Nov How to overcome the challenges in effective real-time CCTV monitoring
It’s natural to believe that with more cameras comes more security, but that is not the case, explains SPA member Mr Eric Brouwers.
There are now over 1 billion video surveillance cameras in the world and that number is increasing rapidly. With this proliferation of CCTV cameras, how can human control room operators realistically be expected to monitor video footage in real-time? It becomes humanly and economically impossible. A massive volume of video is a challenge in itself, but what are the other challenges that the industry faces in delivering effective real-time security video monitoring?
Ineffective and interrupted surveillance
With multiple screens to monitor, control room operators are unlikely to detect less obvious incidents displayed from targets at greater depth or in the background. And when operators do spot something that requires closer monitoring, they must rewind CCTV footage to analyse the incident in more depth.
Often, the operator will have spotted something of little or no relevance. But while they analyse the incident, their attention is drawn away from the real-time feed. That results in frequent temporary breaks in surveillance.
Network bandwidths cannot support surveillance needs
Processing billions of petabytes of video puts enormous strain on servers, while transmitting images from multiple camera locations to physically distant remote-control rooms places severe stress on telecom networks.
These huge data volumes demand superfast data delivery so they can be effectively monitored by control room operatives. This is typically not achieved due to the bandwidth constraints of telecom networks.
Inability to capture non-predetermined threats
Traditional video analytics software relies solely on the predefined situations it has been programmed to look out for. It requires a rule to be defined that specifies the detection of a particular behaviour or event and only then will the software perceive it as a threat.
But what if a threat occurs that is not foreseen or predicted? Most software on the market simply doesn’t have the capability to alert security operators to unpredicted threats.
False positive alerts
There can be many causes of false positive alerts; environmental noise, growing vegetation, moving foliage, wind, fog, rain to name just a few.
Any one of these can alert security operators to what are non-existent threats, particularly in remote locations where camera systems are less frequently maintained. These false positive alerts detract security operator attention and waste valuable time and resources.
Detecting intrusions at night
When little or no external lighting is available at night, thermal imagery is the standard. Thermal imagery provides improved night-time performance but brings the disadvantage of low resolution imagery, limited object classification and significant contrast issues at day-time temperatures.
Can there be a paradigm shift in real-time CCTV surveillance?
We are living in an age of voice-activated assistants and driverless cars. As part of this trend, many video analytics companies have introduced artificial intelligence and deep learning into their software offering.
CCTV cameras are evolving towards IoT (Internet of Things) devices that can be handled by machines rather than by humans, eliminating the need for operators to watch feeds in real-time.
Advances towards faster and more powerful computing technologies, such as multi-core CPUs and graphic processing units (GPUs), will also enable servers and devices to process vast quantities of data in much shorter time frames.
Smart cameras with onboard processing power will become more common, although there is some way to go before their onboard processing power is sufficient to handle very advanced and complex algorithms.
Advances in processing data ‘on the edge’ of surveillance areas, not necessarily onboard the cameras, can also significantly reduce the amount of data to be sent over telecom networks
Video analytics platforms that balance these processing versus bandwidth requirements are much better placed to overcome the constraints that providers are facing today. A layered software structure is the obvious architecture for this, allowing easy scaling of any project.
The next step for the industry is for AI learning capabilities to complement current deep learning technology. This ‘learning’ aspect is still new and only a few suppliers have incorporated it meaningfully into their solutions.
We believe that learning capability is the cornerstone for unbiased qualification in any situation, offering the potential to truly ‘understand’ any scene so that more efficient monitoring can take place. This capability is supported by advanced rule engines to interpret alerts from cameras and to suggest the actions that the operator should take.
There is no doubt that a major shift is underway in real-time CCTV surveillance as technological advances open the door to superior operational efficiency at lower cost.