Camera Surveillance Market Moving Towards the Edge

Over the last 10 years, the surveillance camera market has witnessed drastic changes from analog, closed-circuit cameras, to all-digital IP-based cameras. The physical forms of these cameras have also gone from fixed security-type cameras to wearable body cameras, drone cameras, in-vehicle cameras etc… These changes in the surveillance camera market are brought on by improvements on the hardware side but more importantly on the software side with the recent developments in deep learning (DL). Based on Counterpoint’s estimates, the market will grow to over 200 million units shipped by 2023 at a CAGR of 8%.

Benefits of DL Analytics

AI and DL are playing increasingly more important roles in the surveillance camera market and offer significant benefits over conventional video analytics. Some main benefits of DL are as follows:

Improved Object Detection Accuracy – analytics software can better differentiate between different objects (person, vehicle, animal, etc.), and under various circumstances (day, night, seasonal weather conditions, etc.).

Ability to Learn – traditional Image Video Analysis software relied on a rule-based approach that required configuration by a human operator for each monitoring camera and each type of alert. In contrast, an AI-based solution collects and analyses data over time and creates metadata that describes all objects in each video stream. This technique also allows the solution to scale to an unlimited number of cameras, with no need for a human to configure each new device.

Fast Processing at Lower cost – the rapid increase in processing capacity means that DL algorithms can be run locally at the camera level rather than on a central server or in the cloud. As a result, these algorithms can analyze increasing volumes of video footage in a fraction of the time of earlier analytics.

Surveillance Analytics Moving Towards the Edge

We are seeing a trend emerging where surveillance vendors are moving towards edge-based video analytics applications where data is initially being analyzed and processed on-device instead of on a server or in the cloud. This enables faster response times and results which are sometimes critical for certain surveillance events such as in advanced driver-assistance systems (ADAS). At present, GPU has been the most mainstream DL chip choice for edge-based cameras. For example, Hikvision introduced one of the first AI-enabled front-end products in the market in 2016 – DeepinView binocular intelligent camera containing Nvidia’s embedded GPU Jetson TX1. This GPU is capable of running much more complicated facial, traffic and vehicle analysis and computation with deep learning technology, all running on the device itself without the need to configure a back-end server. While GPU has great computing power, it is also known for having issues in terms of cost, efficiency and power consumption. By contrast, a specially optimized ASIC/FGPA chip for specific vertical applications will have better performance and can be better for inferencing on the edge device. That’s why some of China’s top AI start-ups such as Horizon Robotics, Cambricon, Deepglint and Bitmain, are focusing on developing ASICs for the surveillance sector.

Diverse Applications

There are many applications and use cases for surveillance cameras, software, and different storage solutions, where the focus is generally on safety, reducing risk as well as costs. Assisted monitoring and retail are common verticals where smart surveillance is being applied to. Some interesting use cases are as follows:

Assisted Monitoring – Japanese telecom giant NTT East and start-up Earth Eyes Corp have developed a security camera called the “AI Guardman” which is designed to help shop owners in Japan spot potential shoplifters. It uses open source technology to scan live video streams and estimate the poses of any bodies it can see. The system then tries to match this data to pre-defined ‘suspicious’ behavior. If it sees something noteworthy, it alerts shopkeepers via a connected app. It is claimed that in some early trials the system reduced shoplifting in stores by around 40%.

Retail – No-checkout” retail stores allow customers to walk in, select products and walk out again, without them needing to stand in a checkout line as payment is made automatically through a linked account. This set up can potentially reduce major operating costs for companies as well as address pain points of long wait times for consumers. Amazon launched a “no-checkout” store called, Amazon Go, in late 2016. Alibaba Group also recently opened an experimental cashless supermarket in China, called Tao Café.

Surveillance camera technology has evolved rapidly in recent years as camera prices have dropped and adoption increased. Over the next 18 months, Counterpoint expects that more chipset options will lead to product differentiation increasingly being based on a camera’s analytics software capabilities rather than on the camera hardware itself, although the camera hardware will change and diversify substantially to cater to multiple use cases.