"Machine vision" sounds high-tech, but its logic is simple—using cameras to replace human eyes and algorithms to replace human judgment. On an industrial production line, whether a vision system operates stably depends 80% on the combination of cameras, lenses, and lighting. This article uses the plainest language to help you understand the core components and setup logic of machine vision.
The selection logic for machine vision cameras is completely different from consumer cameras. You might pursue high pixels and nice colors with your phone camera, but on a production line, cameras prioritize speed, consistency, and reliability.
First question: How fast does it need to be?
The production line cycle determines the minimum frame rate of the camera. A line processing 3,600 products per hour handles 60 per minute, or 1 per second. If each product requires one shot, a 30fps camera is more than sufficient. However, for "flying capture" (capturing in motion) on a conveyor belt, you need a higher frame rate and shorter exposure time.
The Shiduwei industrial camera solution based on the OV9281 can achieve 120fps, suitable for high-speed production line applications. For general inspection scenarios, the 60fps AR0135 solution balances performance and cost.
Second question: Global shutter or rolling shutter?
This is the first major decision in machine vision selection. Moving products must use a global shutter—otherwise, the image will distort like jelly. The AR0135 and Python1300 sensors used in Shiduwei's industrial vision solutions are both global shutters, ensuring image accuracy in motion scenarios. If the inspected object is stationary (e.g., printed material sampling), a rolling shutter can reduce costs.
Third question: What resolution is sufficient?
Higher resolution isn't always better. If inspecting a 100mm x 80mm part and needing to detect a 0.1mm defect, you need 100/0.1 = 1,000 pixels in one direction. Considering the Nyquist theorem, multiply by 2, requiring over 2,000 pixels. A 5-megapixel sensor is sufficient. Pursuing excessively high resolution only increases data processing and reduces frame rate.
The lens determines the camera's field of view and working distance. Three parameters are critical:
Practical advice: For simple inspection tasks (presence/absence judgment), an M12 fixed-focus lens is sufficient; for precision measurement, a low-distortion C-mount industrial lens is necessary.
Many beginners spend a lot of time tuning algorithms without realizing the issue lies in lighting. Good lighting design makes defects "pop out," while poor lighting lets defects drown in background noise.
Common lighting types:
Wavelength selection tips: Monochromatic light sources (red, blue, infrared) can effectively suppress interference from colored backgrounds. For example, when inspecting red marks on a green circuit board, using a red LED light source with a monochrome camera maximizes contrast between the mark and background.
Cameras, lenses, and lighting are not selected independently; they require joint tuning:
The success of a machine vision system lies in hardware, not algorithms. A suitable camera paired with good lighting is often more effective than expensive AI algorithms. If you're planning to build a vision inspection system, start with the selection and evaluation of cameras and lighting.
Shiduwei Technology offers industrial camera modules and vision solution selection support. Feel free to inquire.