(480) 630-8816

Hardware Selection for Vision AI

Vision projects implementing artificial intelligence share some basic common components. Camera, computer, power supply and network.

Common Components based on application:

  • Camera for visual data capture.
  • Lens (optional depending on the camera selection)
  • Compute method: Edge Computer, Cloud or Local Server
  • Power and Communications Enclosure. Features may include Cell Module, GPS, redundant power supplies and fiber uplink.

1. Camera Selection

Helpful questions:

  1. Acquisition speed? How fast are the objects moving?
  2. Working distance? How far will the camera be from the area of interest? Will lighting be required?
  3. Camera resolution? How much detail is required for determination?
  4. Environmental conditions? Hazardous materials? Washdown? Temperature.
  5. Lens Selection.

2. Compute Method Selection

Considerations for compute methods include:

A. Processing speed – Do you need real-time notification?

B. System interface – Will communication with other systems or devices such as a PLC be required?

C. Scope – How many inspection points will be required?

Edge: Real-time processing with a single board computer such as a Jetson Nano, Raspberry PI 5 or IPC. Our systems provide a central distribution point for 1-8 POE devices. Smart cameras also have the ability to perform basic AI functions so based on the application the single board computer may not be necessary.

  • CPU Non-Realtime: Raspberry Pi 5
    • Recommended for very low FPS (~1), Hosted Inference, or relaying images.
    • Attributes: CPU only, low FPS, cost effective
  • Realtime: Jetson Orin Nano 8GB – link
    • Recommended for realtime applications (15+ FPS)
    • Attributes: GPU, real-time
  • Multi-Camera Real-time: Jetson Orin NX 16GB
    • Recommended for real-time applications (15+ FPS) of several camera streams
    • Attributes: GPU, real-time

Local Server: On-premise server placed in a central location for scalability and reliability. This becomes cost effective when the number of cameras required supports a larger investment in server hardware.

  • Desktop with 4090: 4090 Lambda Vector One Desktop- link
    • Server Blade with L4: Dell server blade- link
    • Recommended for real-time applications, capable of 100+ FPS
    • Attributes: GPU, real-time, highest FPS
  • Server Blade with L4: Dell server blade- link
    • Recommended for real-time applications of many streams, 75+ FPS
    • Attributes: GPU, real-time, high FPS

Cloud: Flexible and integration-friendly. Many options for cloud processing, service providers supply services and tools help expedite implementation on a subscription basis.

  • Cloud Server: AWS, Azure, Liquid Web
    • Recommended for non real time, higher latency applications
    • Attributes: CPU/GPU, flexible, Cloud Agnostic

3. Power and Communication

Once the above has been determined the power and communications method may selected by asking the following questions:

  1. What type of compute device have you selected?
  2. How many cameras are required?
  3. Are there any additional interface requirements?