With the number of surveillance cameras rising continually and over 566PB of video being stored globally every day, the future of analytics isn't about replacing people, said Mark Wherrett at IFSEC 2018 recently.
It's about helping them to be more efficient, he added on day two of the security show in the Future of Security Theatre sponsored by Tavcom Training.
Wherrett, a tutor at Tavcom Training, said Video Content Analysis (VCA) has been used for many years but can be expensive due to configuring costs and high false alarm rates as well as inaccuracies caused when images captured are of poor quality.
The effectiveness of VCA differs depending on its implementation. For example, while the accuracy of ANPR/ALPR plate reads can exceed 98%, it was recently reported that 98% of face matches on police systems were wrong.
So why the difference?
"VCA works better in a more controlled environment like airports when people are more compliant." Mark Wherrett
"The trouble with facial recognition is that people and their movements are unpredictable which makes their images harder to capture", said Wherrett. "VCA works better in a more controlled environment like airports when people are more compliant."
A good example of this, said the speaker, is when people go through passport gates and comply with the rules set out to ensure the facial recognition works. In contrast, number plates are usually read with specialist cameras and software.
People are more difficult targets compared to cars because they don't follow the same rules of the road, and can cluster together in groups as well as cause shadows that can also confuse surveillance cameras.
How to make VCA work better
According to a report published by IHS, deep learning has been a "revolution in video analytics". This involves teaching systems to become familiar with the same data patterns and can be very effective with good sets of learning examples and works particularly well with facial recognition.
Wherrett highlighted other ways of ensuring that images are captured better, including increasing pixel density, and adding extra pixels on specific targets. It's also important to consider lighting with VCA to avoid issues with shadows and glare.
Most secure systems use multi-spectral imaging like thermal imaging with VCA, visible light and infrared light. All this can make targets easier to identify and categorise for the VCA. Analytics always work best with a broader spectrum of approaches.
Where is investment into VCA being implemented?
"While facial recognition has become increasingly important, facial redaction is also becoming significant for protecting privacy and blurring people or sensitive information that is not of interest in an investigation", added Wherrett. Depending on camera angles used, VCA can automate or at least assist with this.
There is also a significant amount of money going into analaytics deployed in cars. By using multi sensor approaches with multi-spectral VCA, R&D is being carried out with the aim of creating semi and totally autonomous cars.
Video analytics can also be used in fire safety to detect smoke and fire where more conventional detectors can not be used or where the speed of detection is paramount in places such as tunnels or hangars.
Further into the future, the investment in VCA will also enable better monitoring of people who are in assisted living so that they can maintain their independence while guaranteeing their safety.