As we discussed in an
earlier article, multisensor cameras are increasingly equipped with AI capabilities, which can help users in various ways. This note takes a closer look at what types of AI are used, how should the cameras be selected and how they should be installed to optimize performance.
Types of AI used
At the minimum, AI used in multisensor cameras allow for
object detection and classification. This enables the system to detect objects such as people, vehicles, number plates and others.
Then, there are more advanced solutions that can further analyze details in the scene. “More advanced models classify attributes such as age, gender, color and type of clothing, accessories (such as a hat, face mask, glasses, bag) as well as vehicle type, make color and even model,” said Uri Guterman, Head of Product and Marketing at Hanwha Vision Europe.
Advanced AI analytics are able to run on these cameras thanks to more powerful hardware, especially the
system-on-chip, that has enough compute power to execute complex algorithms.
“When processing on the edge, it’s important to have a powerful and flexible SoC that can support multiple analytics processes in parallel while consuming very little power,” said Rui Barbosa, Product Manager at i-PRO Americas, who used his company’s offerings as an example. “Because the field of AI is changing so rapidly, i-PRO has chosen the powerful Ambarella AI SoC which is also advancing the technology behind autonomous vehicles, IoT and robotics.”
Indeed, AI and multisensor cameras form an ideal solution for different types of users, who can be more situationally aware and perform their tasks with more effectiveness. “These capabilities can provide greater insight on what is happening at the site, allowing security operators to make better informed decisions and respond to events. Multisensor cameras have no barriers to using AI analytics compared to standard cameras and the value they bring depends on the use case, the installation and field of view,” said Hamish Dobson, Corporate VP for Enterprise Physical Security at Motorola Solutions.
Selection process
There are certain key considerations to be made when selecting an AI multisensor camera. First, it must be noted that in order for the AI to work effectively, the image quality must good. Therefore, the user should look at the various specs to make sure the camera can produce good-quality image.
“Total
resolution and frame rates are important specifications to consider. Multisensor cameras are intended to be deployed in large areas where people, vehicles and important details might be far away from the camera. The higher the camera’s resolution and frame rates, the easier it will be for it to capture the level of detail required, giving a clearer picture of what’s happening at the site,” Dobson said.
“IR illumination is another important feature to consider, particularly for monitoring outdoor areas where dusk or night time can make it difficult to see. The option to have IR illumination on multisensor cameras allows security operators to see the site clearly even when it is poorly lit or completely dark,” he added.
Being able to work with the right VMS is another key factor to consider. “It’s crucial that the AI camera be paired with a VMS system that supports all the metadata that the camera is generating. If the VMS can’t support all the data from the camera, then the system is limited from the start. This is why i-PRO created the Active Guard plugin that can be installed in popular VMSs such as Genetec and Milestone allowing them to display and work with every attribute the camera can detect,” Barbosa said.
Installation
How to install a multisensor AI camera is also critical if optimal results are desired. In this regard, various considerations also need to be made, for example the area to be monitored, the field of view, and the ideal height for installation.
“For the best AI-based analytic performance, it’s important to not mount the camera too high. A well-placed multisensor camera can easily replace three or four standalone cameras at the corner of a building or a hallway intersection. For hallways, it’s also important to choose a camera that can optimize the aspect ratio to display more hall than wall. Depending on the distance to be covered, it may be useful to specify 4K sensors to ensure the pixel density is high enough to discern faces and other distinguishing features,” Barbosa said.
“Multisensor cameras bring a lot of flexibility and can be installed in many different ways. For example, 180-degree multisensor cameras are typically used for wall-to-wall coverage, whereas 270-degree multisensor cameras are most suitable to cover building corners with no blind spots. On the other hand, 360-degree multisensor cameras offer excellent coverage when installed in the middle of intersections and wide areas such as parking lots,” Dobson said.
It’s good to note that there are certain online tools that can help optimize installation. “Online tools such as the Wisenet Toolbox Plus allow users to make simple field-of-view calculations and select the right capabilities, lens options and accessories for their specific installation. System integrators can easily search, filter and compare product specifications side-by-side and then compile a list of the products required for a specific project. This, in turn, generates a report on the estimated bandwidth and storage requirements for the project,” Guterman said.