Machine Vision in the automotive industry, not just for autonomous cars

// insights

// Electrified

In the automotive industry, stone chips are used as a method to assess the quality of painted surfaces by shooting a determined amount of stone at a surface and then evaluating the extent of the damage. Manual evaluation of the damage can, however, be time-consuming and in addition, there is a risk of assessment errors due to the lack of quantifiable parameters.

I have investigated whether photometric stereo is a suitable method for automatically identifying and analyzing stone chip damage on painted test panels. The focus has been on calculating the extent of the damage and determining how many coats of varnish the damage has penetrated.
Painting is a critical process as it is costly to fix any defects. Since painting is usually a process that is at the end of the process chain, the product has a high value. A lot of material and time has been spent on the product when it is ready for painting, especially in the automotive industry.

Photometric stereo

Photometric stereo is a method that was introduced in the 1980s by Robert J. Woodham (1). The method involves using known light sources together with models for how light is reflected to determine the normality of a surface. Practically, this is done by taking several pictures while illuminating an object from different directions. Then the normal is calculated by comparing the light conditions in the different images. The normal to the surface can then be integrated to recreate a 3D model of the surface.

The project

To investigate whether photometric stereo is a suitable method, a camera rig was developed that can illuminate an object from predetermined angles. Algorithms were implemented for the collection and noise reduction of image data, photometric stereo, integration of gradient fields, identification and classification of injuries.
During the project, it has been shown that photometric stereo is a good tool for this type of problem and there is great potential to replace costly quality controls which are also considered a work environment problem, as it is a monotonous and boring task.

The future

In recent years, increased availability of cheap camera sensors and microcontrollers together with cloud services and good internet communication has enabled cheap collection and analysis of large amounts of data. This combined with great advances in artificial intelligence (AI), which has the strength to be able to analyze large amounts of data and draw complex conclusions, means that computer vision and AI seem to take an increasing role in both our everyday lives and industrial applications. We have seen this development in everything from self-driving cars to Google's AlfaGo beating the best Go players in 2017 (2). Photometric stereo can be an advantageous way of collecting data compared to a regular image because it contains information about the shape of the surface.

For Knightec, it is important to be a driving force in development and by creating broad competence, we can give the customer the best solution. The fact that Knightec is at the forefront of technological development not only ensures our own competitiveness, it also provides an opportunity to help our customers become competitive.

Author

Linus Beccau is an engineer in technical physics at Umeå University with a great interest in technology who during his studies was a driving force in developing Technical Physics Innovatorium, an environment where students can develop their technical interests and build everything from 3D printers and robots to twittering coffee machines. In 2015, Linus won the Technical Physics Robot Competition together with Knightec's Marc Sellgren. Today he works as a consultant at Knightec to develop tomorrow's technology.

References:
1. Robert J. Woodham, Photometric Method For Determining Surface Orientation From Multiple Images. Optical Engineering, 19(1), 191139 (1980)
2. Google AI defeats human Go champion , https://www.bbc.com/news/technology-40042581

Related posts

// Data Driven
// Electrified

Meet Jonas, who automates with machine vision

Jonas Stenvall is one of Knightec's experienced consultants working with machine vision. He is passionate about automating work processes, and in his projects, he works primarily with software development.

Read more
// Data Driven

The potential of Machine Learning

Today, connected devices generate vast amounts of data, purposefully or as a side effect. Those data are precious for a broad spectrum of applications, from analyzing customer behaviour to financial portfolio prediction, making recommendations, categorization and finding outliers. The more data samples that are accessible, the more detailed and correct the result can be.

Read more