printer friendly version

Surveillance infrastructure has changed

The keywords in 2018 for the surveillance industry were AI and machine learning. Marketing departments managed to face-lift the good old notion of neural network and analytics, resulting in a significant increase in demand for facial recognition technologies – in the retail sector, for example.

For hardware vendors servicing the surveillance industry, it turned out to be a challenging period as we have to re-educate the surveillance stake holders regarding the right IT capacity planning. In other words, the processing power required to use AI and deep learning is much higher than companies are used to, and the uniqueness of each site makes it difficult to offer off-the-shelf solutions for server and storage solutions.

The generalisation of complex video analytics is rapidly making the technological model of using general purpose computers or servers in the surveillance business obsolete.

Most of the big-name VMS vendors propose offloading the massive CPU requirement to a GPU (graphical processing unit). If this solution works, the list of compatible GPU products points to high-end graphics cards with a form-factor making it difficult to seamlessly integrate them into existing server/storage solutions.

2019 may result in an intermediate situation where glorified gaming PCs with massive GPU cards will take care of the video analytics while high-end storage/servers will be in charge of the standard video recording. For hardware vendors this is an ideal situation as it is difficult to offer long-term support on hardware produced and designed for the main-stream consumer market.

We will have to answer the demand with that kind of offer at least for 2019, for 2020 there is a remedy coming from the CPU manufacturers: Intel is planning to release a hybrid CPU-FPGA called Xeon-Arria and AMD has patented Zen-based processors with FPGAs on board.

FPGAs are field-programmable gate array chips, in other words, it will be possible for VMS vendors to implement their complex video analytics algorithms in the silicon, resulting in a massive performance boost. FPGA technology has been here for many years, but for some reason failed to become mainstream, unlike GPUs (the gaming market helped in that regards).

Hopefully, if Intel and/or AMD are on schedule, we will see fully integrated server/storage solutions with FPGA accelerated video analytics engines in the surveillance market by 2020.

For more information, contact Franck Martinaux, Capsule Technologies, [email protected], www.capsule-sa.co.za

Share this article:

Further reading: