Ideas maybe nothing new but technological advances are delivering them according to Technical Marketing Services writer Denis Bulgin.
The machine vision industry is extraordinarily dynamic and new techniques and methods are developed on a regular basis. Or are they?
Mark Twain wrote in his autobiography: “There is no such thing as a new idea. It is impossible. We simply take a lot of old ideas and put them into a sort of mental kaleidoscope. We give them a turn and they make new and curious combinations. We keep on turning and making new combinations indefinitely; but they are the same old pieces of coloured glass that have been in use through all the ages.”
Of course, new ideas do evolve at some point, but Twain’s basic premise holds good for most of the recent emerging technologies in machine vision. If we look at some of the latest key topics, such as deep learning and embedded vision, the basic concepts have been around for many years, but advances in technology are allowing them to be turned into practical solutions for real world applications. In recent years, probably best illustration of a maturing vision technology is 3D imaging.
The 3D revolution
3D machine vision is another technique that had been possible for many years. However, creating complex 3D images is computationally intensive. It has taken the emergence of processors capable of handling the computational overhead required for 3D cloud datasets at production line speeds that 3D vision technology became established.
Early in 2012, I attended a 3D presentation event run by a leading supplier of machine vision components and systems. This featured a full day of presentations that covered all of the major 3D imaging techniques and although it was very well attended it was clear that the use of 3D was not yet widespread.
Fast forward to 2017 and at the first UKIVA Machine Vision Conference and Exhibition, the presentations on 3D vision were by far and away the best attended.
During this intervening period, technological developments had allowed much improved performance and accessibility. We saw faster PCs, more sophisticated software for 3D point cloud handling and metrology, and the emergence of 3D smart cameras with on-board processing and measurement.
Improvements in sensor technology and lighting yielded better resolution, but this in turn led to even larger 3D data sets, further increasing the demands on the PC. Today’s FPGA and multicore embedded processor architectures provide faster processing speeds, but now we are also seeing camera manufacturers starting to provide fast, direct memory access between image acquisition and processing on a dedicated FPGA processor before transfer to a PC for further processing. Most importantly, however, over the same period we have seen an explosion in both the actual usage of 3D imaging and the range of possible application areas.
These include general volumetric measurements, completeness checks, part manufacturing inspection, portioning, OCR, distance measurements, packaging integrity and filling inspection, surface finish and many, many more. In particular, there has been extensive use of 3D imaging in robot guidance applications such as pick and place, random bin picking, palletisation and depalletisation and optimising space usage in warehouses.
Coming right up to date, at the 2019 UKIVA Machine Vision Conference and Exhibition, over 25 per cent of the presentations involved some aspect of 3D vision. In addition, the winner of the PPMA’s 2019 ‘Innovative Machine Vision Project’ award had developed a 3D robotic solution for the application of labels to wedges of cheese. Speaking to a major machine vision supplier at a recent engineering exhibition, it transpired that every single enquiry they had received on the first day of the show was related to 3D imaging, even though they had many other techniques on show. Quite clearly over the last seven years we have seen a true maturing of 3D machine vision.
Will the newer technologies emulate 3D?
There are a number of parallels between 3D imaging and the current hot topics such as deep learning and embedded vision. As with 3D imaging, these techniques are not new, but have come to the fore thanks to advances in technology which make them viable for more general use.
For deep learning, massive parallel processing at affordable costs through GPUs, large data storage capabilities and the availability of huge data sets for training have made it a reality. However, in a newer development, inference cameras are now emerging where a trained neural network can be implemented on the camera itself.
We are still at the early stages of implementation of deep learning applications, but there is every likelihood that the technique will follow a similar maturation curve to that of 3D, although the timescales may be shorter.
Embedded vision systems can be viewed in a slightly different way, since they mean different things to different people. Embedded systems are generally considered to be the direct integration of cameras or camera modules into machines or devices using bespoke computer platforms for image processing instead of a classic industrial PC. The best-known example of embedded systems is the smart camera where all of the image processing takes place in the camera itself and this is very mature.
Probably the most exciting development in the field of embedded vision is that of SoC (System on Chip) ARM-based computer technology. This makes it possible to create bespoke systems utilising a wide range of image sensors, standard interfaces and various hardware. With compact designs, simple integration and low power consumption and the increasing move towards integration and connectivity within machine vision and the wider arena of Industry 4.0, the use of SoCs has huge potential, and it will be interesting to see how this develops.
Wherever the latest ideas for machine vision originate, there seems to be a steady pipeline of new technology. It is fascinating watching them mature into established techniques.