How AI Video Intelligence is Teaching Cities to See

Video analytics is undergoing a significant shift. Advances in AI mean cameras are no longer limited to recording events. They can now interpret scenes, understand behaviour, and generate real-time insights.

Traditionally, video surveillance relied on people monitoring screens, reviewing footage after incidents occurred, and manually searching for important events. Today, AI video intelligence is changing that model. Cameras are becoming intelligent sensors capable of understanding situations, identifying behaviours and alerting operators in real time.

We sat down with David Ly, Founder & CEO of Iveda, so he could explain how advances in AI video, natural language search and zero-shot learning are dramatically shortening deployment times while creating new opportunities across smart cities, public safety, retail, transportation and environmental monitoring.

What Is AI Video Intelligence?

AI video intelligence refers to the use of artificial intelligence to analyse video streams and identify meaningful events, behaviours or conditions automatically.

Traditional video analytics typically required organisations to train models extensively before they could reliably recognise new objects. According to David, modern systems are increasingly moving towards zero-shot AI, allowing operators to describe what they want to detect using natural language rather than lengthy training processes.

Instead of teaching a system with hundreds or thousands of examples, users can increasingly provide instructions such as:

• Find a person riding a bicycle
• Detect an e-bike in a prohibited area
• Identify a person wearing a blue hat
• Alert when suspicious behaviour occurs

The significance is not simply technical. It dramatically reduces the time required to deploy new use cases and makes AI accessible to non-specialists.

How Zero-Shot AI Changes Computer Vision

One of the most important themes throughout the discussion is the shift from traditional model training to zero-shot AI. Three years ago, much of the conversation around computer vision focused on training data, model development and accuracy improvements. Today, the emphasis is increasingly on deployment and usability.

As David explains, organisations previously needed large datasets containing labelled examples before AI systems could reliably detect new objects. Modern vision-language models can increasingly interpret natural language instructions and immediately begin searching for relevant objects or behaviours.

The result is faster deployment, lower barriers to adoption, and greater flexibility when organisations encounter new challenges.

Why Ease of Use Is Becoming the Real Differentiator

The discussion repeatedly returns to a simple idea: most decision-makers do not care how the AI works. They care what it helps them achieve.

David argues that organisations often become distracted by technical terminology, while business leaders are primarily interested in outcomes. His emphasis on ‘street-level language’ reflects a broader shift occurring throughout both AI and AIoT deployments.

Natural language interfaces make this easier. Rather than requiring specialist knowledge of model architecture or computer vision systems, users can increasingly interact with AI using conversational instructions. The technology becomes invisible. What remains visible are the operational improvements it creates.

AI Video Intelligence in Smart Cities

Perhaps the clearest examples come from smart city environments. Cities already possess extensive camera networks, but most are still used primarily for recording events rather than actively improving operations. AI video intelligence allows existing infrastructure to become significantly more effective without requiring wholesale replacement.

David raises an example around monitoring prohibited e-bike activity. Rather than assigning police officers to continuously watch specific areas, AI video systems can automatically identify e-bikes operating in restricted locations and generate alerts for the appropriate authorities. The same principles can be applied to parking violations, traffic management, public safety and infrastructure monitoring.

Importantly, the goal is not surveillance for its own sake, but rather to improve city operations, reduce friction, and help cities operate more efficiently.

Related reading: Scaling smart cities with resilient IoT connectivity and Smart Cities: AI & Predictive Modelling Meet IoT.

From Object Detection to Behaviour Understanding

Another major shift is taking place within computer vision itself. Historically, systems focused on recognising objects. Modern AI video platforms are increasingly capable of recognising behaviours and contextual situations.

The distinction is important, as recognising a person and recognising shoplifting are very different challenges.

In the retail example discussed during the episode, AI can identify behavioural patterns that may indicate suspicious activity, such as prolonged loitering, selecting an item, placing it into a bag or pocket, and leaving without approaching a payment point.

This behavioural understanding opens entirely new opportunities across:

• Retail loss prevention
• Public safety
• Security operations
• Transportation
• Smart infrastructure
• Operational monitoring

The conversation suggests that AI video is moving beyond understanding what objects are present towards understanding what is actually happening.

AIoT Applications Beyond Public Safety

AI video intelligence is also finding applications beyond security and public safety. Environmental organisations are exploring how video analytics can help detect smoke, identify pollution events, and monitor protected environments at scale.

Rather than relying solely on manual inspection or post-event review, AI can continuously analyse visual data and flag conditions that require investigation. Similar techniques are being applied to ocean monitoring programmes and even radar-derived imagery, helping experts identify patterns that might otherwise be missed.

These examples highlight how AIoT combines connected infrastructure, AI analytics and operational workflows to deliver actionable insights in complex environments.

The Biggest Barrier to Adoption Isn’t Technology

Despite rapid advances in AI capabilities, David believes the largest obstacle remains awareness and education. Many organisations still view AI as expensive, complex or inaccessible. Others understand the technology but struggle to connect it to their operational priorities.

His argument is that most organisations do not need a technical explanation of neural networks but rather need to understand how AI solves a specific business problem.

What Will Success Look Like?

When asked what the future of AI might look like over the next few years, David does not focus on larger models, faster compute or more sophisticated algorithms. Instead, he argues that success will be measured by improvements to everyday human experiences.

Cleaner cities, faster emergency response, less congestion, safer public spaces, reduced friction in daily life.

The real value of AI video intelligence may not be that machines become smarter. It may be that people spend less time dealing with preventable problems.

Key Takeaways

• AI video intelligence is rapidly evolving from object recognition to behavioural understanding.
• Zero-shot AI significantly reduces deployment time by minimising training requirements.
• Natural language interaction makes AI accessible to operational users, not just technical specialists.
• Existing camera infrastructure can become intelligent sensing infrastructure without wholesale replacement.
• The strongest use cases focus on business outcomes, operational efficiency and public benefit.
• Long-term success will be measured by improvements to human experience rather than technical sophistication.

FAQ

What is AI video intelligence?

AI video intelligence uses artificial intelligence to analyse video streams, identify meaningful events and generate alerts or insights automatically. It moves beyond recording footage to understanding what is happening within a scene.

What is zero-shot AI?

Zero-shot AI enables systems to recognise new objects, situations or behaviours using natural language descriptions rather than requiring extensive model-specific training datasets.

How does AIoT support smart cities?

AIoT combines AI with connected infrastructure such as cameras and sensors, allowing cities to automate monitoring, improve operational efficiency and respond more quickly to important events.

Why is AI adoption still slow?

According to David, the primary challenge is not technology itself but awareness and education. Organisations often need clearer explanations of business value before they invest.

Can AI video intelligence work with existing infrastructure?

Yes. A key theme of the discussion is enhancing existing camera networks rather than replacing them, allowing organisations to gain additional value from infrastructure already in place.

Want to hear the full discussion?

David shares far more detail on AI video surveillance and the future of AI-powered video intelligence in this episode of the IoT & AI Leaders Podcast.

Listen now: https://www.iotleaders.ai/podcast/how-ai-is-teaching-cities-to-see/