Over the past decade, building on years of evolution, advances in analytics running on surveillance cameras have significantly expanded the possibilities to turn data into actionable insights and automated actions. Edge analytics has matured into a highly versatile and innovative field, driven by robust cameras equipped with AI- processing capabilities.
This evolution has paved the way for a new generation of analytics performance. To delve into its current benefits and explore its potential, we spoke with Mats Thulin, Director of AI and Analytics Solutions, at Axis Communications.
The introduction of ARTPEC-9, our latest generation system-on-chip (SoC), marks a significant leap forward in AI-powered edge analytics. ARTPEC-9 enhances object detection and event analysis, while enabling high-quality video with efficient AV1 encoding and robust cybersecurity features.
By achieving advanced, real-time object classification directly within the camera, we are building on over 20 years of experience in developing cutting-edge analytics solutions. Although in-camera analytics is not new, the combination of exceptional image quality and deep learning creates a great platform for boosting performance and analytics capabilities.
Reducing the need for server-based analytics
“Just a few years ago, video surveillance cameras lacked the processing power necessary for advanced analytics at the network’s edge, making server-based analytics, the default choice,” says Mats Thulin. “While server processing power is still widely available, several challenges remain. For instance, compressing video before transmission can degrade the quality of the images being analyzed. Additionally, relying solely on server- or cloud-based analytics when scaling solutions can lead to significant costs, particularly as system demands increase.”
Redefining analytics with AI for unmatched efficiency and precision
AI-based analytics is a highly compute-intensive process. Mats explains: “When analytics run purely on servers, video must be encoded on the edge and then decoded before it can be analyzed, which requires significant processing resources. Managing multiple video streams – even in a relatively small system with 20-30 cameras running at high-resolution and 20 or 30 frames per second – demands considerable processing power.” He continues, “Our edge devices’ processing power, along with their advanced capabilities, has reached a point where sophisticated analytics can be integrated into the cameras themselves. This shift to edge-based analytics reduces reliance on server- and cloud-based solutions, allowing analysis to happen at the point of capture and preserving the highest image quality – a crucial factor for accurate analysis.”
Our edge devices’ processing power, along with their advanced capabilities, has reached a point where sophisticated analytics can be integrated into the cameras themselves.
Images optimized for machines and humans
Advances in image quality consider the fact that images are increasingly being ‘viewed’ by machines rather than being viewed by human operators. This distinction is significant because machines process images differently than humans.
“We can now ‘tune’ video images specifically for AI-based analysis to achieve better analytics results,” Mats explains. “For example, when tuning an image for a human eye”, the goal is often to reduce noise for better visibility. However, for AI analytics, noise reduction is less important.
Improved system scalability with hybrid solutions
Furthermore, analytics on the edge can sometimes significantly reduce the amount of data sent across the network, resulting in greater efficiencies in bandwidth usage, storage, and server capacity. That said, server-based analytics still play an important role. Hybrid approaches that combine the strengths of edge analytics with server- and cloud-based solutions are often the most effective strategy.
Standalone edge-based systems will continue to be valuable for generating real-time rule-based events from object detection and classification. However, Mats emphasizes that, as systems grow more complex, hybrid solutions, which leverage both edge and server analytics, are likely to become the preferred choice for advanced systems.
Unlocking the power of metadata
Metadata – in the context of video – essentially tags key elements within a scene, adding descriptors or intelligence about what is happening, rather than merely capturing raw footage. This additional layer of information enables video management software (VMS) to trigger real-time actions or post-event, targeted searches, and also revealing trends and patterns over time.
With edge analytics, the metadata stream works seamlessly alongside the video feed, optimizing system efficiency. Instead of processing the entire video centrally, only the relevant metadata is analyzed. This approach accelerates processing, enhances accuracy, and reduces costs, unlocking deeper insights without overloading the system.
Protecting privacy with edge analytics
Edge analytics can also help protect privacy. Intelligent masking ensures that sensitive details are obscured at the source. “Faces in a scene can be blurred, and the same goes for license plates or other identifying features,” Mats explains. “If needed, the unmasked video can still be accessed – but only by authorized personnel with high security clearance, and only in specific cases such as an incident.”
Pushing the boundaries of edge technology
We, at Axis, are redefining the possibilities of edge technology with our innovative Camera Application Platform (ACAP). Designed to empower developers, ACAP provides open APIs, robust frameworks, and support for high-level programming languages, transforming Axis devices into versatile platforms for enhanced security, advanced business intelligence, and tailored solutions. "This enables developers to create custom solutions that deliver outstanding performance and scalability", Mats explains.
And as technology evolves, Mats highlights, "Axis will stay at the forefront, equipping developers to shape the future of smart edge applications across industries like retail, healthcare, urban planning, and more."
Moreover, Axis is supporting ONVIF metadata specifications to foster an open ecosystem. This approach allows developers to create hybrid architectures with ONVIF-compliant products, ensuring seamless system integration and interoperability across diverse environments.
Shaping smarter, more robust systems across industries
The evolution of detection and classification is transforming industries across the board. “Advances in object detection and recognition, with cameras gaining a deeper understanding of their surroundings – distinguishing between a street, a lawn or a parking lot – enable more precise analytics, explains Mats. “This level of semantic segmentation enhances scene understanding, drives innovation, and fosters more robust systems.”
Edge analytics are reshaping industries by bringing AI-powered analysis directly to the source, which unlocks new efficiencies and innovations. With improvements in detection, metadata integration, and scalability, its potential to transform operations across a wide range of domains is clear. Mats Thulin puts it simply: “Edge analytics processes data where it’s generated, opening up new opportunities for smarter, more responsive systems in numerous fields.”