Improving people's safety with AI assisted video surveillance
Evolution is a constant and occurs to cope better with an ever-changing environment. Biologically, this can be associated with Darwin’s theory of evolution and regarding manufacturing and production, the Industrial Revolution offered a critical turning point to drive efficiencies and increase product yield on a monumental scale. More recently, say over the last 50 years, computer systems have delivered many efficiencies for numerical calculations and statistical analyses. Without these technology advances, analyses and calculations would have to be conducted manually which is time consuming and a mundane task for the human brain.
Fast forward to today and the talk of AI (Artificial Intelligence) and Machine Learning is “top of mind” for many execs. Commonly discussed is: How can one harness vast quantities of data leveraging AI to gain advantage; whether for increased safety, fraud detection, better products, or simply to avoid humans having to undertake repetitive activities?
Moreover, two pivotal factors have converged to drive industry adoption. Firstly,: the complexity of AI algorithms is now so sophisticated they can mimic (and often improve) on what a human brain can do. Secondly: the associated hardware “horsepower” to support these complex AI algorithms is now affordable. These 2 factors combined allow many industries to deploy AI with a tangible ROI.
One area which is gaining notoriety is how to detect events from video footage, also referred to as Computer Vision or AI Assisted Video Analytics. Video surveillance usually involves one of more operators sitting in front (if they are lucky) of a high-res multi-screen video-wall looking for anomalous activities. The not so lucky, stare at a single screen (often split into 4 feeds) which rotate on rolling basis depending on the camera count. As you can expect, events get missed! The human brain is just not equipped to do this 24x7. People tire, need breaks, have lapses in the concentration, and can only focus on a small number of concurrent tasks. How can they be effective at monitoring what can often be tens to hundreds of camera feeds and how can this mundane human task be improved?
Rather than being reliant on a human eye(s) detecting events of interest (which as discussed is often missed) computer vision can do to do this instead. This enables 24x7 coverage on several concurrent camera feeds using a system that never tires and can evolve with changing circumstances. Now the ethical question: Does the computer/system instigate an action having ID’d an event? In most cases, no. With surveillance, particularly human surveillance, any detected events can be passed to an operator who can then decide an appropriate course of action. This results in the operator becoming more effective. Or, a single security person is able to monitoring several hundred concurrent camera feeds. It is the role of the AI stack to determine what characteristics or events are of importance. Some characteristics that can be considered for detection include people recognition; geofencing; anomalous event detection; perimeter breach; weapon detection; sentiment analysis; exclusion zone activity to mention but a few. All these events and their detection is key to contribute to a safer environment particularly where people are concerned.
Hewlett Packard Enterprise (HPE) has a series of Edge designed systems capable of powering the aforementioned video-walls and AI stacks. These Edge systems are designed for placement in remote or branch locations. They are smaller, more efficient and more durable than datacentre designed systems; and are better suited for Edge deployments. For video surveillance, Edge compute delivers instant detection and notification of events. As it relates to surveillance and people’s safety, latency is key. A delay in detection can
jeopardise life and thus Edge compute with AI assisted surveillance adds significant value where human safety is of paramount importance. For more information on HPE’s Edge capabilities please refer to:
- https://www.hpe.com/uk/en/servers/edgeline-systems.html
- https://assets.ext.hpe.com/is/content/hpedam/documents/a00074000-4999/a00074055/a00074055enw.pdf
