Using Artificial Intelligence to Improve your CCTV System
As artificial intelligence is beginning to permeate our daily lives, we take a look at how it impacts CCTV security systems.
11 October 2018
As artificial intelligence is beginning to permeate our daily lives, we take a look at how it impacts CCTV security systems. We begin by examining how artificial intelligence (AI) is been developed at Netwatch and then we’ll explore the direction other industries are taking with this technology.
Let’s first take a step back and look at the technological leaps in CCTV technology that have allowed us to get to this point:
An Internet Protocol (IP) camera is a type of digital camera that can send and receive data via a computer network and the internet. Unlike their traditional CCTV predecessor (analogue closed-circuit television cameras), IP cameras can transmit high resolution footage which can be viewed remotely.
Video Camera Analytics is the capability of automatically analysing video to detect and determine temporal and spatial events. There are different functionalities within Video Camera Analytics. One of the more simpler forms is where motion is detected in relation to a fixed background scene. Camera analytics form the basis of how AI works in CCTV technology.
AI utilises software programs that analyse the images from cameras in order to recognise humans, vehicles or objects that display out of pattern behaviour. In addition to simple rules, such as restricting humans or vehicles from certain areas at certain times of day, more complex rules can be set. For example, the user of the system may wish to track the direction of vehicles e.g. monitoring vehicles entering, rather than leaving a premises.
This is the most prevalent type of AI within the security industry. In rule-based intelligence, a programmer sets the rules for the system and the system responds only to those rules.
Active Artificial Intelligence
In active artificial intelligence, AI learns what is normal behaviour for people, vehicles or machines based on the environment and its own observation of patterns. When it observes an anomaly it sends an alert. For example, the system would know that it is normal for cars to drive in the street, so if it observed a car driving up onto a sidewalk it would flag that as an anomaly.
How we currently use AI
Netwatch Software is focused on rule-based AI. We currently use AI to increase the capabilities of camera detection and to improve operational efficiency.
Increasing camera capabilities
Products like our loiter detection are based on multiple predetermined rules set up by our engineers. This program will allow individuals in its field of view to walk through camera zones, but will trigger an alarm if the individuals loiter longer than a predetermined period of time.
Our Hotspot detection works in a similar way, whereby using a thermal analytics camera, we can detect heat and if the detected temperature exceeds the predetermined point an alarm is triggered in our Communications Hub.
Improving alarm handling efficiency
The speed at which we receive and action an alarm is paramount in safeguarding our customers’ assets. CARS is our proprietary alarm handling software. It ensures that once the camera analytics triggers an alarm that it will be verified by an intervention specialist within 30 seconds. Watch the video below to see how our software works.
Machine learning and spiders
Spiders have a fondness for nesting near and around security camera lenses, restricting the field of view for the camera or triggering a false alarm. Using machine learning, we can feed a software program with a bank of video clips, both with and without spiders, for detailed processing and analysis. The ideal result will be that the AI will learn what a spider on the camera lens looks like. This will then trigger a notification for the client or service engineer that the camera needs to be cleaned. This improves alarm handling efficiency (by cutting down on false alarms) and ensures that the security system is not compromised by even the smallest intruders.
How companies are developing AI for the security industry
Quick search using natural language
Ella is a reactive focused CCTV system which provides consumers with a useful google type search feature to quickly recover footage using natural language. Using tagged metadata and deep learning algorithms, Ella will allow you to recover relevant clips using a convenient search feature. For example, if you searched “Fed Ex on Friday” it would show the clips relevant to that search. This use of reactive AI technology is based on the traditional model of recording crime, rather than preventing it. A more proactive system would detect the crime before it happens and intervene based on the target behaviour in time to prevent it happening in the first place.
Machine learning and calculation of footfall
Camera-based footfall analytics are generally installed over entrances to retail stores. They are also used in airports, train stations and other public places. However many more cameras are required for larger spaces. Footfall analytics can provide accurate and detailed information on footfall counts and provide other useful items such as heat maps and dwell times where enough cameras are installed. Knowing where customers move is extremely useful for retailers. It allows them to optimise their store layout and know where to place merchandise e.g. popular vs. unpopular, expensive vs. cheap items. This tool can add real value to an existing CCTV systems.
People Counting, Age and Gender Statistics
Panasonic has adopted this deep learning technology in its face recognition products for the security industry. This has led to the successful development and commercialisation of revolutionary face recognition technology which overcomes the difficulties of conventional technologies e.g. recognising faces when they are tilted, changed by aging, partially hidden by sunglasses etc.
Other applications include:
- VIP lists – make staff aware of important individuals (VIPs) and respond in an appropriate manner
- Black lists – identify known offenders or register suspects to aid public safety
- Banking transactions – verification of the persons attempting a financial transaction
- Access Control verification – confirming identity visually, manually or automatically
The age of true AI machine learning systems will soon be upon us. Many applications will go unnoticed to the average consumer who will not realise that their product or service is being enhanced using AI. The evolution and landscape as to where companies are choosing to develop AI continues to change, so it’s not yet possible to buy an off the shelf true AI powered CCTV system. When a true AI system is capable of improving operational efficiency or saving money on security costs expect to see a wider rollout.
Most people don’t tolerate risk when it comes to security, so until we can rely on the results coming from AI we will continue with the more tried and tested rule-based intelligence programs.