EUROPEAN MACHINE VISION FORUM
image. Te discriminator then takes the adapted images and a real image dataset, and tries to classify which is real and which is fake. Over time the generator gets better at producing more realistic images, while the discriminator becomes more adept at flagging synthetic data. Tis process produces a realistic image. Tere are two branches of GAN: the pixel-
level approach, like Cycle-GAN, and the feature- level method. Pixel-level approaches don’t exploit any semantic information, i.e. the context of the image, so a framework like Cycle-GAN can introduce a lot of artefacts, such as trees in the background. Zama Ramirez worked with a new pixel-
level GAN approach that exploits semantic information during the generation process. Here, the discriminator does not only classify if the image is real or fake, but also performs
The advantage of training a neural network on synthetic data … is that [labels are] almost for free
semantic segmentation of the image. Tis leads the generator to produce images that have the same semantic content as the source synthetic images. Te group worked with a dataset of 20,000
training images with semantic labels from Grand Teſt Auto V, and a validation training set of 2,975 cityscape images without labels. A network was trained on GTA adapted images, and then the performance evaluated against the cityscape validation set. Te performance of the network trained on
Pirelli to improve quality control with automated tyre inspection
Italian tyre manufacturer Pirelli has built a prototype inspection system to improve quality control. Te Automatic Visual Control
(CVA) project adds automated inspection and digitisation to Pirelli’s tyre manufacturing processes. Pirelli has around 30,000 employees and 20 production plants around the world. Vincenzo Boffa, head of the
vision technology group within the R&D department at Pirelli Tyre, in Milan, presented the CVA project at the European Machine Vision Forum in Bologna, Italy, in September. Boffa explained that automated
surface inspection of a black opaque tyre is not easy to achieve. In addition, a tyre has a large surface area when both the external and internal surfaces are taken into account. Tyres are complex objects,
made up of different layers and different compounds. A ‘green’ tyre is built up of layers of semi-finished material and then moulded and cured to produce the final product.
Boffa compared tyres to
pharmaceutical products, in that the quality of the end-product is important for Pirelli’s integrity. A lot of instrumentation is used to control the stability of the layers and compounds, and Boffa explained that manual inspection by employees is also a big part of the quality control checks that are carried out in production. Te aim of the CVA project
is to add automated inspection to the manufacturing process. Digitisation of the data is also important, as the information can be used to avoid defects in future production runs. ‘If we find a small defect, which is not a true defect, we can understand how to improve our process, how to avoid these defects in the future,’ Boffa said in his presentation. Automated inspection also reduces the amount of repetitive work for employees. Te prototype system is made
up of three robotic islands equipped with 2D and 3D cameras. Te tyre is rotated on its axis on a rotating table and is inspected with the cameras – the
14 Imaging and Machine Vision Europe • October/November 2018
GTA adapted images increased from 18.23 per cent to 31.4 per cent mean intersection over union (mIoU), and from 60.43 per cent to 80 per cent pixel accuracy, compared to just using GTA raw synthetic data. ‘Training a network on our adapted images
can achieve almost double that from training a network on just synthetic data,’ Zama Ramirez commented. ‘Te adapted images belong much more to the real distribution than the synthetic images.’ However, he added that the adapted
images ‘still can’t reach the same accuracy of performance as when trained on real data.’ Te group is now employing a Cycle-GAN
approach, consisting of two generators and two discriminators to try and achieve even better performance. Whether Grand Teſt Auto can be made to appear completely real is yet to be seen.
tyre is automatically centred using a vision system. Te inspection data include a
3D profile of the circumference of the tyre made by laser line imaging. Pirelli also uses a technique that illuminates the tyre with three different light sources: diffuse light, and leſt and right grazing light. A line scan camera, operating at one frame every 60ms, then captures three images under the different lighting conditions, to identify very small defects, such as small cuts. Te system uses high- resolution cameras, along with
customised, powerful LEDs. Pirelli makes thousands of
different tyres, of sizes ranging from those for 16-inch rims to 24 inches, and up to 30 inches in the US. Te size of the tread can also vary substantially. ‘Our machine has to adapt to all the different sizes of tyre,’ Boffa explained. Te project was conducted in
collaboration with the University of Bologna, Specialvideo, an Italian industrial vision system designer, and Politecnico di Torino. Pirelli has filed 43 patents relating to the CVA prototype as of the end of 2017.
@imveurope
www.imveurope.com
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72
