Traditionally, a person was required to physically stand and supervise a machine and its procedures. At a logistic warehouse, this meant employees would need to check the label of packages and put them in the right locations manually. Nowadays, all these tasks can be automated using machine vision.
“Machine vision can be characterized as capturing visual data from the environment, interpreting it to make sense of if, and deciding what actions to take,” said Jerome Gigot, Senior Director of Marketing at Ambarella.“This is conceptually similar to human vision -- the eye captures visual data from the environment, the brain interprets it, and finally the brain decides whether to act on it.”
Machine vision systems, particularly those using cameras linked to multiple edge computing servers located inside a factory, can be used to automate certain industrial processes. This includes routine inspections, quality assurance, process control, predictive maintenance, safety inspections, inventory management and barcode reading.
Basic components of a machine vision system include a processor, a camera and a camera interface to digitize images. With the help with other components, such as light sources, lenses, image processing software, object-detection sensors and input/output hardware, a machine vision system is complete.
“In a machine vision system, the sensor subsystem is the ‘eye,’” said Gigot.
This “eye” consists of a lens to focus the light on the image sensor, and an image sensor converts this light into digital data, or “pixels,” which presents different colors and luminance data.
Some machine vision applications, such as barcode scanning, use only black and white images, since color information is not required to make an intelligent judgement for these types of functions. In some other cases, such as identifying paint defects, color information is critical.
Previously, digital images collected by machine vision systems were sent back to a PC or an FPGA (Field Programmable Gate Arrays) back-end system for analysis. By placing this video processing capability directly into the camera, the time taken to identify images is reduced, speeding up the whole production process.
“With the latest advances in chip technology and the advent of dedicated silicon to process vision data -- such as the latest series of Ambarella CVflow system-on-chips (SoCs) -- this processing can be done in the camera itself, right next to the sensor,” said Gigot.