Contributed by Jonathan Haslanger, Solutions Architect at SparkCognition.
Visual AI is the next generation of video analytics technology, helping you maximize the information and value you can realize from your original investment in camera infrastructure. But what are the best cameras for visual AI systems?
As a solutions architect at SparkCognition, I spend a lot of time talking to customers about configuring visual AI systems for their organization. One of the most common misconceptions I’ve observed is the assumption that cutting-edge solutions like SparkCognition Visual AI Advisor require fancy cameras to leverage the benefits. Customers want to know if they will need to replace their camera system. I’m always glad to assure them that this is not the case!
In fact, Visual AI Advisor requires nothing more advanced than today’s widely available and inexpensive GPU nodes—it will work with virtually any camera system feed available on your site, including CCTV, PTZ, drones, and mobile devices.
With that said, it’s wise to consider camera suitability for AI applications when upgrading or deploying new cameras, even if you aren’t shopping for an AI solution today. I’m always glad to help our customers evaluate their options from a cost vs. benefit perspective.
Camera selection for Visual AI: If you can see it, AI can see it, too
As a rule of thumb, remember this overarching advice: What helps a person see also helps the AI see. In other words, whatever camera configuration makes it easier for human eyes to observe the action will similarly help Visual AI Advisor view events within the field of view, as well.
When customers ask me for guidance in selecting cameras for visual AI systems, I recommend several factors to keep in mind:
- Pixel density
- Field of view
- Lens type
- Frame rate
- Low light performance
- Environmental considerations
Let’s take a closer look at each of these factors.
Pixel density, a measurement of resolution at distance, is crucial for AI models, as they require a minimum number of pixels on the object being tracked. To determine if a camera can provide sufficient resolution for your desired AI use case, calculate its resolution at distance in terms of pixels per meter (PPM). For use cases where the camera detects smaller items (such as cigarettes or cell phones), a higher PPM is required than when larger items (such as vehicles) are detected. Consult with your AI provider for recommended PPM for your specific use case.
Field of view is another important factor. Ensure that the activity you want to track is well-contained within the camera’s field of view and not blocked by obstructions. This includes both permanent obstructions (such as trees or buildings) and temporary ones (such as vehicles or crowds). Lenses that distort the image around the edges (like fish eye lenses) can limit the effective area for AI in the field of view. Additionally, moving cameras (such as PTZ cameras) are not ideal for continuous monitoring with AI, as they can be pointed away from the planned field of coverage, resulting in gaps in coverage. Instead, I prefer using modern multi-lens cameras that can cover a wide area with consistent high-resolution coverage.
Frame rate is also important for advanced AI systems that derive insights from understanding movement. Cameras supporting a frame rate of 15 frames per second or better are preferred. However, lower frame rates may be sufficient in use cases where objects are moving slowly.
Low light performance is another crucial factor to assess per your environment. Logically, if the camera cannot “see” the scene, the AI will not be able to detect anything. Some cameras have IR or low-light modes, but their detection range may be reduced by up to 50% in low-light conditions. Options to overcome low light include artificial lighting, which can be deployed to improve scene lighting and camera detection range.
Finally, consider any environmental considerations when selecting cameras for AI use. For example, cameras mounted outside may require an IP rating to be water/weather-resistant. In hazardous environments, cameras may require special certification or NDAA compliance.
By taking these factors into account when selecting cameras for visual AI use cases, you can ensure that your camera system will optimally support the features that make SparkCognition Visual AI Advisor so impactful.
A solution for all camera types: SparkCognition Visual AI Advisor
SparkCognition Visual AI Advisor makes it easy to leverage proven enterprise computer vision technology to automate real-time visual analytics and alerts using your existing camera infrastructure. It requires no new camera equipment to deploy, scales to thousands of cameras in days, consolidates all video feeds into a single interface, and ships with 125+ proven use cases.
Did you know that Visual AI Advisor was just announced as a 2023 product winner in the computer vision category? No other computer vision-enabled product today combines the breadth of features and enterprise scalability offered by Visual AI Advisor:
- Pairs with any camera system already installed on your site, including CCTV, PTZ, drones, and mobile devices.
- Deploys securely on-prem and/or syncs with your cloud using our low-code/no-code integration framework.
- Automates visual analytics on multiple real-time video streams at once and tracks your alerts on a single user interface for all cameras and use cases.
Want to learn how organizations apply Visual AI to predict and prevent accidents and near misses, improving worker safety, productivity, and satisfaction? I highly recommend our on-demand webinar: 5 Ways to Transform Worker Safety with Artificial Intelligence.
If you have any questions or want to schedule a demo for your team, reach out today!
