Deep learning is taking video analytics far beyond what previous analytics could do. It’s not just a case of more accurate analysis or more specific types of object recognition. As a result, preparing to deploy apply video analytics requires preparation far beyond what was needed for previous generations of video analytics.
Q: Our organization is making use of AI deep-leaning in a variety of ways, and I keep getting asked if Security is considering modern deep-learning analytics. What should we be thinking about?
A: There are several very important steps involved in considering the effective use of emerging deep-learning-based video analytics.
Preparation
Making use of today’s emerging deep-learning-based video analytics requires much more preparation than for previous physical security technologies. It goes far beyond simple product or cloud service selection. Key elements to consider are:
- Security technology infrastructure readiness
- Extent and nature of existing video coverage
- Discovery of new business benefit opportunities
Over the past 15 years, my company has been called on to troubleshoot many security video analytics deployments that were not living up to expectations. Analysis of those and other troubled analytics projects found that incomplete preparation in the areas listed above was at the root of their shortcomings.
The points listed above are important for any size company – but grow more important the larger the organization is and the more facilities it has. For small companies, the stakeholders are just a few and their collaboration is easy to incorporate into normal business planning activities. The larger the organization is, the greater the number of stakeholders to whom security video analytics is relevant. This complicates strategy development, stakeholder engagement, and scoping the application of video analytics.
This column briefly looks at each preparation category. Further explanations will be found in an upcoming Security Business magazine article about “AI readiness.”
Security Technology Infrastructure Readiness
If an up-to-date video system software and hardware inventory don’t exist, create one that includes information on hardware (device make and model, firmware versions, warranty and support life) and software (installed version, the latest release, warranty and support life). It also helps to document the installed cost of video systems plus annual service and maintenance. It provides a financial context for support and service costs, as well as total investment insight.
Many network cameras outlive their end of support life, meaning that from a cybersecurity perspective they are not safe to continue using if firmware updates will come to an end.
The nature of the technology underlying security systems has evolved drastically over the past two decades. Download the PDF diagram titled “Advancing Physical Security Technology” to rate where your own electronic physical security technology fits into the picture. It provides a context that can help in explaining the status of current security technology to non-security decision-makers.
AI deep-learning-based video analytics require much more processing power for their massively parallel processing than any previous security system products have needed. See my recent article about the newly released AWS deep-learning technology for video analytics, which involves special technology developed for cloud computing, on-site appliances and built-in on-camera capabilities.
Existing Video Coverage Landscape
For most medium and large organizations, security video system infrastructure has been acquired site by site over the past decade or more, resulting in a mix of camera fields of view selected based on earlier technology limitations rather than what’s best for the risks being addressed. Furthermore, facility physical security risks and related corporate liabilities have increased, resulting in video coverage shortcomings.
An assessment of current video coverage should consider these factors:
- Are enough cameras in enough places to detect and document potential situations of security concern?
- Are the camera fields of view acceptable?
- Is the detail in each camera image sufficient for the purposes of the camera, such as activity detection, person or object recognition, and person or vehicle identification?
If camera coverage is insufficient, adding video analytics won’t fix the existing camera deployment shortcomings.
Many video deployments make use of motion-based recording and motion-based alerting and alarming.
- To what extent is motion-based alarming in use or should be used?
- What portion of motion-based video alerting are actual alarm situations? Is (or would be) the rate of nuisance alerts or alarms excessive?
Many organizations don’t make extensive use of motion-based alerting because camera or VMS masking isn’t flexible enough to allow the full exclusion of no-interest motion. This is a case where AI-based video analytics can help remedy that shortcoming, and many existing security deployments could benefit.
For example, for organizations monitoring a large amount of video motion alarms, a high-value recently-emerged technology is the cloud-based Calipsa system, a deep-learning false alarm reduction application that typically achieves between about 80% and 95% false alarm reduction using only three images of scene motion activity spaced one second apart. Most existing network cameras and video encoders of can email just three still images or a 3-second clip or to Calipsa without burdening existing networking and Internet services. No additional on-site hardware is required.
Discovery of New Business Benefit Opportunities
As has already been proven in the retail business sector, video analytics can provide information that is of high value to business operations and planning. This is an area for thoughtful research and internal exploration. Today’s deep-learning-based video analytics can increase the number of business stakeholders for security video, with the result that there are often more security video analytics stakeholders outside the security function than within it. For example, in pharmaceutical manufacturing facilities, there are often more cameras in use for product quality and plant safety purposes than for security purposes.
Much to Consider
Deep-learning based facility video analytics products are still in their infancy. Although their ability for false alarm reduction and inferring the risk potentials of activity patterns based on what they continue to learn is very amazing now, their capabilities will continue to advance at an increasingly high rate. This requires a deployment strategy that involves a continual review of expanding AI capabilities, against the backdrop of facility activity context and the value of information that the analytics can glean about it.
The deployment of deep-learning-based analytics is a more complex undertaking than for previous types of security system technologies. It can require significant forethought and preparation to ensure the satisfactory results will be achieved for the investment and that an improving ROI based on technology advances is part of the picture.
About the author: Ray Bernard, PSP CHS-III, is the principal consultant for Ray Bernard Consulting Services (RBCS), a firm that provides security consulting services for public and private facilities (www.go-rbcs.com). In 2018 IFSEC Global listed Ray as #12 in the world’s Top 30 Security Thought Leaders. He is the author of the Elsevier book Security Technology Convergence Insights available on Amazon. Mr. Bernard is a Subject Matter Expert Faculty of the Security Executive Council (SEC) and an active member of the ASIS communities for Physical Security and IT Security. Follow Ray on Twitter: @RayBernardRBCS.
© 2021 RBCS
