Making machine vision an enabling technology


Neil Sandhu, Chairman of UKIVA (the UK Industrial Vision Association), explains how machine vision has evolved into an enabling technology across the manufacturing sector.

Machine vision is a fast-moving technology. It has continuously been able to take full advantage of the latest developments in semiconductors and, therefore, has become a truly enabling technology for the manufacturing industry. Regular improvements bring better performance for existing applications and open new possibilities in a variety of market sectors.

The growing combination of vision and robotics enhances opportunities to automate manufacturing processes. From the early days of Windows 95-based PCs to the latest parallel processing architectures, ever more powerful computer processing has supported the evolution of machine vision capabilities. This is true for computationally intensive techniques such as 3D imaging, hyperspectral imaging, and deep learning. At the same time, improvements in image sensor technology have led to cameras with higher resolution, faster operation and smaller physical size, as well as systems that make use of light outside the usual visible wavelength range.


Harnessing processing power

The development of PC-based image processing toolkits and libraries has been crucial in extending the range of machine vision capabilities. These have created familiar tools for a host of vision tasks such as: detecting the presence/absence of features, part-alignment, part-measuring, surface inspection (looking for scratches, stains or irregular features), 1D and 2D code verification, character reading and label inspection. As processing power increases, more tools become available to expand inspection possibilities further.

Another major benefit is establishing previously specialist techniques such as 3D imaging, hyperspectral imaging, and most recently deep learning, as accessible and affordable options. Now that 3D point clouds can be manipulated and analysed with similar ease to 2D image data, 3D vision has become a core component of machine vision with roles in quality inspection, robotics and depth perception. Hyperspectral imaging combines infrared spectroscopy with machine vision to locate differences in the chemical composition of organic materials, offering major new possibilities for detecting impurities in various industries. Deep learning techniques can solve imaging problems that traditional machine vision methods cannot easily address by training convolutional neural networks (CNNs) to recognise features or defects in images for classification purposes and then applying this to new images. This computationally intensive process frequently uses the parallelisation offered by the latest graphic processing units (GPUs) in PCs.


Smart cameras and embedded systems

Smart cameras are embedded technology devices with onboard image acquisition, processing and analysis capabilities. They can perform inspection tasks without an image processing PC and send the inspection result directly to the process control system using industry-standard communication protocols. They offer great versatility, particularly where few cameras are required and can be reset for different product types and even moved to different locations to carry out different inspections. Many use an ‘app’ approach to enable specific measurement apps to be run on the camera. As onboard processors capable of handling 3D calculations emerged, 3D smart cameras also came to market. Most recently, smart inference cameras have become available where trained neural networks can be downloaded to the camera to enable deep learning inference to take place directly within the camera. Vision systems can also be based on embedded, single board platforms such as NVIDIA Jetson, Raspberry Pi, CompuLab and ODROID to allow the creation of self-contained vision solutions. An even newer approach is the system on chip camera, which is optimised for advanced digital imaging combined with a comprehensive image processing library.


Keeping up with the trends

The vision industry has developed several dedicated data transfer hardware interfaces to accommodate the greater volumes of data that accompany many of these developments. These include CameraLink, CameraLink HS, GigE Vision, USB3 Vision and CoaXPress. Keeping up with the technology and trends across the industry can be challenging, especially during current times. To address this, UKIVA is providing two broad-based information platforms: the online Machine Vision Conference Technology Presentation Hub 2021 (, which goes live on 15 July, and an interactive roundtable webinar featuring six vision industry experts which will take place on 20 July (


About Author

Comments are closed.