printer friendly version

Adding edge intelligence to surveillance

As we have seen in many articles in this year’s SMART Surveillance Handbook, adding intelligence to surveillance infrastructures has become critical. AI, in the form of identifying people or other objects, stopping false alarms from inundating the control room, and warning of potential security or operational problems is available in different forms today.

Some cameras have built-in AI capabilities, most management platforms offer AI analytics on the server, and cloud-based solutions abound. The edge is also a popular destination for running AI in order to prevent the transmission of large amounts of video footage, as well as reducing false alarms. While the edge can mean AI in the cameras themselves, it can also mean hosting servers or some device in branch offices where a site’s surveillance footage is analysed before an alarm is raised.

Some edge devices are NVRs, which save the footage and have enough processing power to run analytics, others are different, stand-alone devices specifically designed for intelligent functionality. Of course, the efficacy of these devices depends on the processing power onboard, the quality of the AI labelling (or training), as well as the design of the software to be resource friendly without letting any events slip by.

Whichever solution is chosen, the customer and users will need to be attached to some management platform that is able to send and receive alerts (apart from the video management platform) and enable operators or users to take appropriate action.

A low-cost tower on the edge

Taking all these issues into account, local company Sentronics has developed and released edgE:Tower, an intelligent video surveillance agent built on deep learning AI technology. The edgE:Tower device is basically a low-cost computer designed purely to apply AI technology to video samples and return the relevant information to users. The building blocks of the solution, according to Bernard Senekal, MD of Sentronics, are IoT, AI and Blockchain (which is used to assist in training and validating object classifications).

edgE:Tower is linked to the free Telegram messaging platform, making it easy to receive alerts and send instructions to the device. Users can allow one or more people to manage the platform with different permissions.

The system has been built on NVIDIA Jetson hardware and the software is designed to offer reliable AI functionality that can scale. The company’s developers have optimised the code so that it can offer the full AI experience even on entry-level NVIDIA kit.

The AI model is trained for object classification (people, vehicles etc.), along with direction and counting features, as well as posture detection (for example, picking up duress). Avoiding the trap of consuming large amounts of bandwidth, the system needs a 480 x 360-pixel image to make accurate long-distance classifications – the models have been trained to ‘understand’ what they are seeing at between 120 and 180 metres.

Senekal explains that a 10-channel unit costs about R10 000 and is ready to run out-of-the-box (the edgE:Tower does not rely on customers to train the AI). So, for about R1000 per channel, companies can obtain a 95% or more reduction in false alarms, while being alerted in real time to various real events. Additional options include a 16-channel system expandable to 20, and a 32-channel system expandable to 40. A 128-channel solution will be released later this year.

All that is needed on the customer’s side is the edgE:Tower device, power and an Internet connection. All the system requires is a RTSP video stream and once plugged in, you are ready to go. Interoperability with other platforms is also possible via an open API.

How it works

When the edgE:Tower detects a real event that should be dealt with, the user/s receive a notification via their Telegram app. The app itself has a number of options they can then choose, from activating defences (like sirens or lights) or notifying a security company. The notification comes with a 7-second video clip to show what is happening, and this can also be incorporated into the reporting functionality of the system.

An important aspect of the edgE:Tower AI is accuracy, not only in false alarm reduction, but in not missing a real alarm. While reducing false alarms is important, nothing is as important as not missing a single real alarm event. In a test run, against another solution offering similar features, with one of its corporate sponsors, out of 1000 alarm events, edgE:Tower only let 24 false alarms through compared with 80 from its competition. Critically, edgE:Tower did not miss any real alarms, while the competitive product missed 143.

Nothing is as good as a demonstration. Readers are invited to go to https://t.me/et_edgetower_demo_bot and sign up for a demonstration (you will need to be on your mobile and have the Telegram app installed). You will be able to play with various alarm scenarios and the options on the app to react to those alarms. Other examples are available at https://edgetower.co.za/gallery.

Flexibility and evolution

The key to edgE:Tower is flexibility. While designed as an edge solution, customers can also implement it as a cloud product, or even a hybrid solution. Furthermore, Senekal says the solution is trainable to ‘see anything the eye can see on video’.

Looking ahead, he says that the company is working on additional features and solutions to run on edgE:Tower, including a retail analytics suit. Additional work is being done and we will see more AI in the near future, such as facial detection and LPR, as well as anomalous scene detection.

The focus of edgE:Tower has been to develop an accurate and reliable AI solution at a low price point with negligible deployment and maintenance costs, and Senekal believes this has been very successfully accomplished.

For more information, contact Sentronics, +27 82 497 0897, [email protected], www.sentronics.co.za/www.edgetower.co.za

Share this article:

Further reading: