MACHINE VISION

What is Machine Vision?

Machine vision (MV) is a complex multidisciplinary activity requiring an in depth knowledge of optics, illumination, electronic devices, algorithms, computer software, AI, system design, project management and most importantly a correct understanding of the clients process and requirements.

Machine vision systems use a single or multiple acquisition devices to acquire images of the client’s products. These images are first processed and secondly interpreted by computer vision algorithms. The resulting data or decisions are then used to instruct other systems in the client’s process how to act.

Acquisition devices may be 2D or 3D in nature. 2D cameras may be colour, monochrome, visible, NIR, SWIR, UV or thermal. 3D devices may be based on structured lighting techniques (laser profilers, grid projectors etc), time of flight devices or stereo vision.

Some typical MV applications: Precision part location and orientation, precise measurements, defect or part identification in extremely fast processes, robot guidance, pick and place, bar codes, colour tones, multi spectral, 3D etc.

Deep Learning

Deep Learning (DL) extends the range and the types of applications which can be handled by Machine Vision. It is ideally suited to classifying objects or defects in repetitive and pseudo repetitive patterns or textures no matter how complex. The key to developing a successful DL application is the availability at design time of an extensive set of images which fully represent all the possible variations in the process to be inspected. In addition these images must be acquired in conditions identical to those were the system will be deployed. These images are used to train, fine tune and verify the performance of the Deep Learning.

Some typical DL application sectors: Wiring harnesses, mechanical assemblies, fabrics, ceramic tiles, moulded glass, plastic films, solar panels, wood, printing, OCR, food products, medical pills, welding seams, product appearance identification, aesthetic defects in parts etc.

A hybrid approach

Machine vision and deep learning can be used in a hybrid manner within the same system. A correct implementation maximizes the strengths of each technology while at the same time minimising its inherent weaknesses.

Why?

Quality control

Prevent out of tolerance production

Reduce returns from clients

Objective and repetitive inspection

Inspection of 100% of the product

Parts identification

Measurement control

Robot guidance

Process automation

Thermal and spectral information cannot be readily seen with the human eye

Kentec s.r.l.

Sede Operativa:
Via Boccioni, 2 56037 PECCIOLI(PI)