In video surveillance, edge computing refers to the processing of data in the camera instead of in the cloud. Today, more and more manufacturers have rolled out the so-called “AI camera” to suit the edge computing trend.
In video surveillance, edge computing
refers to the processing of data in the camera instead of in the cloud. Today, more and more manufacturers have rolled out the so-called “AI camera
” to suit the edge computing trend.
The types of analytics
that can be supported by the AI camera are quite diverse. “These include human face, physique, gesture, posture, movement, gender, age, voice and language analytics, as well as analytics that can identify objects, architecture, food, plants and animals,” said Jonny Wu, VP of AndroVideo
The benefits of running these algorithms at the edge device are manifold. “Edge computing allows instant response as the camera itself does the detection and recognition; there’s no need to transmit the video for analysis in the backend. And because of this, bandwidth consumption is also minimized,” Wu said. “Moreover, since video can be stored on the edge as well, the cost of cloud storage is saved. Edge cameras are much easier to install, compared to setting up servers and the associated software. Finally, people are always concerned about privacy issues, and edge cameras that are running locally can alleviate people’s concerns.”
As an example, AndroVideo’s own AI camera supports facial recognition, pose estimation, human detection and age/gender recognition. “The facial recognition is performed on the camera. Once the camera detects a blacklisted individual or someone who is not supposed to be at a venue, it will store the associated image and text in the SD card for the end user to access,” Wu said. “Further, if the user so chooses, the camera can send an alert to an alarm device as well as the user’s mobile device over the internet.”
Components-wise, the processing unit and the image sensor both play a key role. “The processing unit, whether a central processing unit, graphics processing unit, neural network processing unit or tensor processing unit, decides how powerful and sophisticated the AI model can be hosted,” Wu said. “A good image sensor, meanwhile, can capture good-quality video at high frame rates, even in poorly lit scenarios, to enhance recognition.”
Applications of AI cameras are wide-ranging. “A non-exhaustive list would include safe city applications, such as fire, traffic and policing; gaming; corrections; healthcare; education and utilities,” said Dave Staudacher, Senior Software Architect at Pelco
“Ultimately, every sector will benefit from increased ‘intelligence’ in video surveillance systems. Automated actions triggered by live data based on historical understanding – for instance warning audio announcements in response to loitering or suspicious behavior – will aid security, while more rapid response to accidents will improve public safety,” said Mats Thulin, Director Core Technologies at Axis Communications
Beyond security, more and more these AI cameras can be used for business intelligence purposes as well. “AI cameras … can help retail organizations increase their competitiveness and commercial success. For example, VIP customers who opt to participate in special marketing promotions or other incentives can be identified so staff can provide the right kind of service at the right time,” said Max Fang, IP Project Product Director at Hikvision
. “In the same way, AI cameras can help to better understand customers and meet their needs more effectively – leading to more commercial opportunities.”
“Outside of many security benefits, AI systems integrated with cameras could deliver on a wide number of requirements for an organization, such as automatic restocking of shelves in retail outlet as customers select items and place them in their shopping cart. In an office environment, when an employee removes supplies from a stock room replacement items could be immediately logged for reorder and keeping required stocking levels low,” said Jeff Whitney, VP of Marketing at Arecont Vision Costar
. “AI could interact with building systems, adjusting heating and cooling without human intervention as the number of people increases or decreases in specific areas of a building, or operate blinds as outdoor lighting varies, as seen by the camera systems in place.”