Exclusive: Spotlight on AI-based Edge Analytics

AI

Share this content

Share on facebook
Facebook
Share on twitter
Twitter
Share on linkedin
LinkedIn

Chris Bishop, Sales Director APAC & Marketing Director at Ipsotek, an Atos company, explores the latest analytics technology.

In today’s world the use of CCTV cameras in security, surveillance and health & safety applications has become ubiquitous with 770 million cameras installed worldwide according to a report by Comparitech. With some analysts predicting this number will increase to one billion by the end 2021. So how do you monitor hundreds or thousands of CCTV cameras connected to a wide area or wireless network to detect incidents or suspicious behaviour reliably?

In this article we will explore how AI-based Edge Analytics built on the Atos Computer Vision Platform offers scalable solutions for multiple vertical sectors including Transportation (Rail, Aviation, Roads/Highways, Ports), Safe/Smart Cities, Critical National Infrastructure, Government, Retail & Logistics and Energy.

Today, 95% of data is non-analysed and 80% of this non-analysed data consists of images and videos. Analysing this complex and voluminous data in real-time can be time consuming and complicated.

Computer vision is a process that involves capturing, processing and analysing images and videos to allow machines to extract meaningful, contextual information from the physical world.

To process these large volumes of complex data in real time, Atos Computer Vision leverages GPUs (graphics processing units) in its BullSequana Edge servers. These powerful compact servers are ideally suited for edge deployment and reduce the need to stream IP CCTV cameras to a data centre for video analytics processing.

It is forecast that within the next two years, 99% of video and image content captured for enterprise applications will be analysed by machines. Computer vision will enhance human capabilities by assisting in everyday tasks and keeping track of multiple, real-time events.

Atos Computer Vision Platform

The Atos Computer Vision Platform delivers the power of server-based processing with AI applications at the edge. IP camera streams are processed by the BullSequana Edge or BullSequana Edge nano servers which use the NVIDIA GPU to provide real-time alerts (metadata and event images) based on pre-trained AI models and Scenario-based Rules.

Metadata related to a detected event is stored in a database node and both the metadata and event images are sent to the Video Management System (VMS) or Command, Control and Communications system (C3) for review and management by the operators according to the organisation’s Standard Operating Procedures (SOPs). With existing integrations to many of the world’s leading security manufacturers’ VMS and C3 systems, the Atos Computer Vision Platform and BullSequana Edge series server architecture fits neatly into new and legacy systems.

Using the BullSequana Edge series analytics architecture the volume of data sent from the edge server to the VMS or C3 is significantly reduced, as data is only sent when an incident or alert is generated as shown below.

Network ConnectionBandwidthDescription
Management Node → VMS / C3MinimalEvents
Management Node → VMS / C31 MBEvent Image
Management Node → VMS / C340 KbpsEvent Metadata

In May 2021, Atos acquired Ipsotek and now uses VISuite, which forms the core of the software toolbox, referred to as Atos Computer Vision Platform. Within this highly scalable platform there are five software modules that can be operated stand-alone or integrated with a VMS or C3 system.

Five software modules

VISuite AI – Artificial Intelligence, generates real-time alerts from existing or customised AI models and scenario-based rules and actively monitors large networks of cameras and tracks objects in real-time throughout the FOV of the camera(s). Advanced trackers and AI detectors are used to maintain a track on every individual, vehicle, or object to generate rich and accurate metadata unique to each object in the scene.

Through VISuite’s deep learning capabilities, bespoke neural networks can be trained to detect novel objects on a project-by-project basis.

VISuite FR – Face Recognition utilises advanced face recognition algorithms to detect faces even in crowded scenes. Precise, accurate biometric information is created in the form of a unique FIR (Facial Identification Record) which can be stored and compared with a database of images for multiple applications. In addition, VISuite FR can detect people not wearing face masks in designated zones and has the capability to pixelate an individual’s face to protect their privacy. Another application is able provide a second level of access control verification for securing premises or sites by associating a registered face with an access control card.

VISuite FR actively monitors large networks of cameras and captures faces in real-time throughout the camera FOV. Advanced trackers and AI detectors are used to maintain a track on every face. Incident Response is an ergonomically designed GUI (IR GUI) that allows an operator to upload a face and search through hours of video in seconds. Allowing a person to be tracked throughout the CCTV network.

VISuite LPR – License Plate Recognition, utilises advanced Licence Plate Recognition (LPR) algorithms to automatically detect number plates even in a busy scene. Precise data created from the reading of number plates can then be stored or compared to a database for use with multiple applications. The watchlist solution detects plates that have previously been enrolled in a watchlist and raises an alarm when a match is found. It is also possible to calculate the average speed of a vehicle by measuring the time taken to traverse a certain distance.

VISuite LPR can read a number plate at a certain point in the road and start a timer. The LPR engine will recognise when the same car passes the second camera and from the time taken to travel this known distance, calculate the individual vehicle’s average speed.

VISuite Incident Response – IR GUI is an intuitive user interface that allows operators to define an event and then search through hours of video in seconds, across multiple overlapping and non-overlapping cameras to track a “tagged” individual or vehicle. Person(s) of Interest (POI) can be tracked across the network of cameras in real-time with their path overlaid on a map.

VISuite Protect – represents the operational benefits associated with deploying the five VISuite components of the Atos Computer Vision Platform. Efficiency, productivity, accuracy, reliability and privacy.

Real world applications

If we consider all the components that comprise the Atos Computer Vision Platform what types of applications are there? Within transportation, the railway sector has many applications. A typical train station has multiple areas that must be monitored 24/7 as shown in the indicative diagram below.

The BullSequana Edge server could be connected to cameras across the station and adjacent areas while the Bullsequana Edge nano could be connected to cameras covering remote areas such as level crossings or maintenance sidings and facilities, with secure WiFi or 5G backhaul links for the Metadata and Events to the Management node. The configurations and settings for the BullSequana Edge series servers would be managed via a central management node server in the command-and-control centre to ensure compliance with operational rules and regulations and the management of user profiles.

This edge-based architecture offers organisations maximum flexibility with cameras connected to the servers via conventional networks or secure wireless. If new cameras are required, they are connected to the edge server which sends meta data and event images to the VMS or C3 for review and actioning instead of streaming all the cameras across the existing network.

In this example, it is also possible to have multiple stations each with their own local edge servers processing CCTV footage, connected to a central monitoring centre (CMC) which may be in a different geographic location. The additional network bandwidth required between each station and the CMC is negligible, as the Atos Computer Vision Platform alerts and metadata are relatively small compared to the CCTV camera IP streams. This federated architecture allows changes to be made to one area of a station, whilst maintaining existing settings at all other stations.

AI

Solutions for all of the sub-categories listed in the diagram above are available as standard within the Atos Computer Vision Platform. As is the ability to create customised dashboards showing dynamic data for control operators. This example shows footfall on a platform during peak travelling hours and the corresponding heatmaps. The data to create this type of dashboard is available within the DB node and is accessed using Kibana. The mix and match of data to be displayed is flexible and customer determined.

As the number of installed CCTV cameras reaches the one billion mark, along with the development of new smart cities, airports, railways and other interconnecting infrastructure, the need for reliable and scalable solutions for the automated analysis of video and images will continue to grow. The Atos Computer Vision Platform offers an End-To-End solution for AI-based Edge Analytics that is tried and tested with small footprint servers that are energy efficient offering reduced carbon emissions and lower operational costs.

For more information, visit: www.ipsotek.com

This article was originally published in the December 2021 edition of International Security Journal. Pick up your FREE digital edition here

Newsletter
Receive the latest breaking news straight to your inbox