How Citilog develops applications for Axis network cameras through ACAP
Axis is a company that truly believes that “the whole is greater than the sum of the parts”. Through the Axis Developer Community – which is open to all developers, whether already working within an Axis partner company or not – and the Axis Application Development Partner (ADP) Program, the company provides a wealth of resources which connect the ecosystem around Axis products and technologies. In doing so, and particularly in giving early access to new technology, innovation is accelerated, and connections are made that bring benefits to partners and customers alike.
Within the ADP Program and through the Axis Developer Community, partners and developers gain access to the Axis Camera Application Platform (ACAP), which specifically allows for development of applications that sit within surveillance cameras themselves (and an increasing number of other products). As the capabilities of Axis surveillance cameras increases – and particularly as cameras are now available which include a Deep Learning Processing Unit (DLPU) – ACAP represents a place where some of the most cutting-edge innovation is taking place.
A fundamental change to the business
The partnership between Axis and Citilog, a specialist in advanced traffic management analytics applications, began in 2009, at around the same time that ACAP itself was created (and well before Citilog was acquired by Axis in 2016).
Jean-Marie Guyon, Vice President at Citilog, explains how the company immediately saw the opportunity in ACAP when it was first announced: “Citilog had already been an Axis partner for a number of years when the capacity of the camera processors became sufficient to port our applications on the edge. But, as soon as we saw the possibilities for developing analytics applications that were integrated into surveillance cameras themselves, we knew it would fundamentally change our business, even if it took longer for customers to realise the potential!”
Decreased need of bandwidth with analytics on the edge (in the camera)
Prior to the ability to develop in-camera analytics applications – often known as ‘edge analytics’ - the analysis of video took place on centralized hardware and servers, either housed within the customer’s own premises or within a data center. This meant the transfer of huge amounts of video footage from the camera to the data center, with an associated demand for bandwidth and the inevitable cost.
Jean-Marie continues: “We immediately saw the opportunity to remove a significant proportion of the bandwidth demands through in-camera analytics. Put simply, by analyzing the video within the camera, we only need to transfer the data that matters rather than everything. For cameras that are monitoring roads 24 hours a day, seven days a week for incidents that can be relatively rare, it’s obvious to see what a difference this could make.”
Market slow to respond
But as Jean-Marie mentions, while Citilog’s developers immediately saw the potential, the market was, as usual, slower to respond: “Over the past decade we’ve done a lot of promotion and evangelizing of the benefits of in-camera analytics. It takes time for sectors to change and adapt new approaches – and not least when it requires a change in hardware – and the first few years were tough going. However, persevering has been worth it, and today more than 70% of our business is based on ACAP. More than that, in the past few years we’ve seen the majority of tenders demanding in-camera analytics.”
On a step-change in capabilities of in-camera analytics
So much for the past, however. At its heart ACAP is a platform for innovation, and Citilog is always looking towards the future. Deep learning represents the next area of innovation. Jean-Marie expands: “With the combination of our current deep learning-based solution (CT-ADL: Citilog Applied Deep Learning) and the evolution of the in-camera processing capabilities, it feels like we’re on the cusp of a real step-change. With the AXIS Q1615-LE Mk III, we have the first Axis camera in the market that includes a deep learning processing unit (DLPU) which combined with the CT-ADL makes it the first operational DL-based solution running on the edge. It’s difficult to overstate the scale of the step forward that this represents.”
Deep learning (DL) is a subset of Artificial Intelligence (AI). In (very) simple terms in relation to video analytics on the edge, the primary benefits relate to much greater accuracy in the detection, identification and classification of all types of object: a generic ‘vehicle’ becomes a car, lorry, bus, or motorcycle and, critically, objects that aren’t relevant can be safely ignored. But deep learning is something that requires huge amounts of processing power, and while this was previously only available through the use of remote servers, it’s now accessible in the camera itself.
One major benefit: Reduced number of false alarms
In a sector such as traffic management, the ability to differentiate between a greater number of objects is critical, as Jean-Marie explains: “One of the biggest issues for any surveillance operation is the cost of false alarms: alerts being triggered that require attention and prompt action, but which aren’t actually materially important. In traffic management, as an example, traditionally one of the most common causes of false alarms are shadows and rain puddles; these can often be mistaken as a vehicle, and if they’re in the fast lane of a motorway, will create an alert. The power of deep learning reduces these false alarms substantially. In fact, that’s an understatement: we’re typically finding that the number of false alarms is reduced by a factor of 10.”
Another benefit: Reduced need for hardware
Such a substantial reduction in false alarms is one obvious benefit to customers, but the switch from processing power in the server and data center to the camera also means a reduced need for hardware. These ‘lighter’ solutions are therefore easier to maintain and further reduce operational costs. They also open up new use cases for in-camera analytics where lack of available bandwidth would have previously made it impossible.
While Citilog’s focus remains focused on traffic management – and through that using the analytics and data created to deliver on the vision for smart cities – the potential for deep learning edge analytics is there for every industry sector and use case. Harnessing the creative power of the largest number of people has always been at the heart of the Axis ethos. Through the Axis Developer Community, Axis ADP and ACAP, the opportunities for all developers and partners to learn, experiment and innovate are infinite.
Access tools to develop ACAP applications