AI and machine learning for enhanced video surveillance security – A balancing act?

The integration of artificial intelligence (AI) and machine learning (ML) into physical security systems is rapidly transforming the video surveillance industry. Dakota Murphey provides an overview of the opportunities and precautions the security industry must consider as adoption rates in video surveillance grow.

From smart alarms to CCTV camera monitoring solutions integrated with intelligent analytical algorithms to analyse video footage and detect anomalies in real time, the surveillance landscape is rapidly seeing evolution at the hands of AI/ML technology.

Not only this, but AI-powered facial recognition software provides a whole new dimension to biometrics and access control.

This isn’t even mentioning the amount of data that can be autonomously aggregated and condensed to inform top-level strategic decisions – as far as surveillance system integrators and providers are concerned.

Is the correct and ethical use of AI in video surveillance a careful balancing act? (Image credit: Brain light/AlamyStock)

In short, it’s clear that AI/ML tech in video surveillance provides numerous promises, from enhanced security to greater productivity and actionable insights collated more quickly. As a result, it may be evident to begin integrating more AI/ML programmes and tools into your operations, however, you should be mindful of certain risks and limitations when adopting this technology at scale.

The promise of AI in video surveillance

• Automated Monitoring: AI allows surveillance systems to continuously and autonomously monitor environments for abnormalities and threats. The consistency and high level of accuracy allows security teams to dedicate more resources and time to higher-level tasks. Machine learning algorithms can be trained to identify signs of trespassing, loitering, vandalism, violence and fire.
• Real-Time Alerts: Intelligent cameras with built-in analytics can immediately notify security personnel when a high-risk event is detected, allowing for rapid response. This is far more effective and efficient than manual video reviews that are subject to human error.
• Facial Recognition: AI facial recognition provides a touchless biometric solution for visitor management, access control and watchlist screening. Stadiums, airports and other high-traffic facilities are employing this technology to identify individuals of interest for swift containment and referrals to authorities if needed.
• Crowd Analysis: Computer vision algorithms can scan crowds to detect high-density levels, suspicious abandoned objects, abnormal noise levels, higher-than-normal temperatures, smoke and many other hazardous scenarios. This allows officials to manage crowds proactively and avoid situations from escalating into dangerous ones.
• Investigations: AI enables intelligent video search to easily surface clips of interest from vast surveillance archives. Security staff can use visual data mining to quickly investigate incidents and find correlations.

Concerns around AI-powered video surveillance

While the benefits of widespread AI adoption in a security or surveillance environment seem apparent, unsupervised use can yield some concerns for both integrators and end users.

• Cyber security: Connected camera networks – along with incumbent enterprise infrastructure – create plenty of attack surfaces vulnerable to compromise by malicious hackers. Steps must be taken to encrypt video feeds, install software updates, and follow cyber security best practices. This includes prioritising continuous software patching, stringent access controls, encryption, MFA (multi-factor authentication) and investing in off-site penetration testing solutions to identify hidden vulnerabilities.
• Privacy: Many organisations have raised concerns about the privacy implications of widespread public camera systems coupled with technologies like facial recognition. There are fears about constant monitoring and tracking of individuals without consent, along with suspected unscrupulous use of this data to be given to mysterious third parties.
• Bias: Like any technology, AI models reflect the unconscious biases inherent in their training data. Critics point out that facial recognition systems often perform worse on women and people of colour. Therefore, measures must be taken to ensure fair, unbiased usage and continual improvement.
• Over-reliance: There are worries that over-dependence on AI surveillance could lead to complacency and substandard human oversight. There are also concerns that AI/ML adoption could put workers at risk of employment termination if organisations look at the bottom line; the technology should augment security staff, not replace their expertise.

Balancing innovation with responsibility

The risks highlighted above should not deter the adoption of AI in surveillance outright, but rather compel companies in the industry to deploy the technology responsibly and not at the extent of human supervisors and operatives.

In short, while security firms advocate the application of AI to help security teams do what they already do, albeit faster and with greater accuracy, there is still a need for responsible human input.

With careful oversight and governance, AI/ML can make monitoring environments more secure and alleviate security personnel from low-level tasks that can instead be entrusted to computers and algorithms.

A balance of healthy AI/ML integration with resource optimisation can help companies adopt a more secure and less time-intensive process with teams fulfilling more high-value work.

Here are some other best practices to follow:

• Conduct thorough, conclusive assessments on privacy and data protection before deploying organisation-wide facial recognition, profiling, and aggregation technologies. If end-users express concerns, be forthcoming and transparent about the use and monitoring of data, with clear and unambiguous policies.
• Train AI models on diverse, unbiased datasets to avoid inaccurate or discriminatory results. Rigorously test systems for fairness defects and ensure that any known flaws are rectified promptly and without prejudice.
• Clearly define the division of task management between AI and human workers. Maintain the ideal amount of experienced security operatives and personnel and upskill them – if necessary – on casting judgement, overriding errors and validating decisions that AI programmes make.
• Develop strict policies and safeguards around data retention and lawful use that align with regulations like GDPR. Give individuals visibility and control over their personal data.

The future of AI-enabled surveillance

AI promises to unlock safer, more data-driven security and monitoring capabilities than ever before. However, thoughtful governance and diligent oversight are imperative as these powerful technologies continue evolving and proliferating.

The video surveillance industry must lead by example, pioneering AI applications that enhance security while still upholding privacy, accountability and fairness.

Listen to the IFSEC Insider podcast!

Each month, the IFSEC Insider (formerly IFSEC Global) Security in Focus podcast brings you conversations with leading figures in the physical security industry. Covering everything from risk management principles and building a security culture, to the key trends ahead in tech and initiatives on diversity and inclusivity, the podcast keeps security professionals up to date with the latest hot topics in the sector.

Available online, and on Spotify, Apple Podcasts and Google Podcasts, tune in for an easy way to remain up to date on the issues affecting your role.

Listen to the podcast today