11 Jun From video streams to actionable security data

As artificial intelligence transforms surveillance from passive observation to proactive insight, Walter Candelu, Vice President EMEA, Real Networks, explores how video analytics is reshaping security operations.

For decades, surveillance systems have operated as passive observers. Cameras captured vast amounts of footage, but their role was largely limited to recording events for later review. When an incident occurred, operators would rewind hours of video to reconstruct what happened. The system was reactive by design — useful for forensics, but limited in preventing incidents. Video analytics fundamentally changes this model. By applying artificial intelligence — particularly computer vision and deep learning — video streams can now be transformed into structured data. Instead of raw footage, systems generate metadata: objects detected, behaviours identified, events flagged.

This transformation is not incremental; it is architectural. Video is no longer just content — it becomes a data source. Once in this form, it can be processed, correlated and integrated in ways traditional surveillance systems never allowed.

From reactive surveillance to proactive and predictive security

Turning video into data enables a shift in how security operations function. In a proactive model, systems detect events in real time. Unauthorised access, loitering, wrong-way vehicle movement, or unusual crowd behaviour can trigger alerts instantly. Operators no longer need to monitor dozens of screens continuously; the system highlights what matters.

The next step is predictive security. By analysing patterns over time — movement trends, behavioural anomalies, recurring risks — systems begin to identify deviations before they escalate into incidents. While still evolving, this represents the direction of travel: from monitoring events to anticipating them. The real value of video analytics lies not only in detection accuracy, but in its ability to shift security operations from response to foresight.

Video analytics as a core sensor in the security ecosystem

As video becomes data, it naturally integrates into the broader security ecosystem. Other systems — access control, intrusion detection, traffic management — have long operated on structured data. They generate events and logs that can be aggregated and analysed within command-and-control platforms. Video, by contrast, has traditionally been the most disconnected component.

Video analytics closes that gap. Camera systems can now produce events that correlate with other data sources. A badge swipe can be matched with a detected face. A perimeter alert can be classified and filtered in real time. In this model, cameras evolve from passive recorders into intelligent sensors within a connected system. They contribute to a unified operational picture, enabling faster decisions and more coordinated responses.

The reality of deployment: Key challenges for video analytics

Despite clear technological progress, deploying video analytics at scale remains complex. While algorithms have improved significantly, the surrounding ecosystem — tools, infrastructure and standards — has not matured at the same pace.

Three challenges consistently emerge: a fragmented vendor landscape, limited adaptability across use cases and the difficulty of handling large-scale data processing. These are not theoretical constraints, they are the main reasons many deployments stall after initial pilots. Understanding these challenges is essential to moving from promising technology to operational reality.

A fragmented market and the challenge of fit

The rise of artificial intelligence has lowered the barrier to entry for video analytics, resulting in a crowded and fragmented market. Startups, software vendors, camera manufacturers and integrators now offer solutions that often appear similar, but vary widely in performance, scalability and robustness.

For security professionals, this makes selection increasingly complex. Evaluating solutions goes beyond feature comparison. It requires understanding how models are trained, how they perform in real-world conditions, how they scale and how they integrate within existing systems while meeting data governance and privacy requirements.At the same time, no single solution fits all use cases. Security needs vary significantly across industries; from crowd analytics in transportation to perimeter protection in logistics or loss prevention in retail. Each environment requires different models, thresholds and operational logic, yet many platforms still offer generic capabilities that only partially address these needs.

Customisation is possible, but often introduces complexity, longer deployment cycles and higher integration effort. Organisations must therefore navigate both a fragmented vendor landscape and a persistent gap between standardised solutions and real-world requirements.

From proof of concept to real deployment: when systems break at scale

Scale remains the most significant challenge. Video analytics generate large volumes of data that must be processed, transmitted and analysed in real time.

While small deployments can perform efficiently, scaling to hundreds or thousands of cameras — producing tens of millions of events — quickly strains bandwidth, processing power and system latency. Centralised architectures often struggle under this load, becoming complex and costly to maintain.

The challenge is not only video, but the volume of events and metadata that must be managed and correlated effectively. Many solutions perform well in proof-of-concept environments, but real-world conditions expose weaknesses. Variability in lighting, weather and camera placement, combined with network limitations and increasing data volumes, can significantly impact performance.

This gap between pilot and full deployment is where many projects fail. The technology may work, but the system is not designed for scale. Bridging that gap requires a focus on efficiency, robustness and realistic deployment conditions from the outset.

Edge processing and the rise of smarter cameras

One of the most promising developments addressing scalability is the shift toward edge computing. Instead of processing video centrally, analytics can be performed directly on cameras or nearby devices. Advances in hardware have enabled this shift. Camera manufacturers are integrating more powerful chipsets capable of running AI models locally. This reduces the need to transmit high-resolution video streams, lowering bandwidth usage and improving response times.

Edge processing enables more scalable architectures by distributing computational load. It also improves system resilience, allowing analytics to continue functioning even when network connectivity is limited. This trend marks a critical step towards making video analytics practical at scale.

The industry’s opportunity: Standardisation and interoperability

As systems evolve, the next major opportunity lies in standardisation and interoperability. Today, different manufacturers implement analytics capabilities in different ways. Data formats, APIs and integration methods vary across platforms, making it difficult to build cohesive, multi-vendor systems. This fragmentation increases complexity and creates the risk of vendor lock-in.

Standardisation offers a clear path forward. Interoperable systems would allow organisations to integrate technologies more easily, adapt to changing requirements and scale without rebuilding infrastructure. The opportunity now is for the industry to align around common frameworks; enabling more open, flexible architectures and unlocking the full potential of video analytics within a unified security ecosystem.

What comes next for video analytics

Video analytics has moved beyond experimentation. The core technology is real, and its capabilities continue to improve. The trajectory points towards increased edge processing and more robust, scalable and adaptable systems. The question is no longer whether video analytics works — it clearly does.

The challenge now is turning that capability into an interoperable and reliable source of data that can operate effectively in complex, real-world environments. That transition — from innovation to operational reality — is where the future of video analytics will be defined, and where camera systems will fully take their place within broader situational intelligence platforms.

www.safr.com