logotipo ISR
Buscar
Cerrar este cuadro de búsqueda.

Difference between linear and matrix systems

Tabla de contenidos

Linear and matrix systems

Anyone involved in the field of computer vision is well aware of the current trends and how the sector has evolved in recent years. Convolutional Neural Networks (CNNs) and Transfer Learning have become part of our daily conversations, allowing us to process images with hundreds of different approaches. Edge Computing is becoming indispensable in increasingly demanding applications with volumes of information to process that grow exponentially year after year. Big Data techniques applied to the data generated by automatic inspection systems enable us to draw conclusions that transform industries. However, we must never lose sight of the foundation of computer vision, which is image acquisition and the proper industrialization of image acquisition technology.

When a company asks for help considering introducing automatic inspection into its productive process, the key points are the system precision and its payback. Focusing on ISR’s automatic aesthetic inspection systems, a well-acquired image where defects are highly visible, and inspection technology adapted to the production will make it possible for to effectively help this company.

At this point, in our quest to align the inspection system with the production process, we will make an initial division of the production processes based on whether discrete parts are produced (e.g., any component used by various TIERs as well as OEMs to assemble a complete vehicle) or if a continuous material is produced (such as a coil of steel, plastic, fabric, or paper).

Differences bewtween linear and matrix systems

Image acquisition available technologies

When the public thinks about image acquisition, they imagine taking out their phone, opening the camera app, and snapping a selfie while enjoying a pleasant moment with family or friends. To immortalize the moment as accurately as possible, users seek a mobile device with a camera that has the highest number of megapixels, meaning a higher resolution. As an engineer who analyzes and appreciates units, I would like to pause on this aspect. We are talking about an image being composed of pixels, and a sensor capable of simultaneously measuring the values of tens of millions of these pixels, which are then organized into rows and columns to form that desired image (for simplicity, we will skip topics such as RGB channels and Bayer color compression).

This simultaneous capture of pixel intensity values across all the rows and columns that make up an image is what we associate with matrix technology today. Matrix sensors are those capable of simultaneously capturing all these pixels, resulting in a straightforward, complete capture.

As an example, this image capture process largely adapts to the production process of a discrete component, such as the transparent covers of headlights and taillights described in the latest blog entry. These pieces will be placed within an inspection system and analyzed by capturing the necessary images.


Although we have successfully implemented automatic inspection within the first type of production process listed above, deeper reflection leads us to the conclusion that this matrix technology will be challenging to integrate into a continuous production process due to its continuous nature—you cannot stop the material!


The sensor (in other words, the camera) needs a specific amount of time to acquire the information that will compose the image. This is known as «exposure time.» If the object or surface being captured moves during this time, it will result in a blurry final image. This is the reason why, when taking a photo of an athlete, arms and legs are not well defined. If this happens in an inspection system, it would fail the first premise we established previously: «defects must be clearly visible in the image».

Let us make some calculations in order to clarify this topic. The typical exposure time for a matrix image in industry ranges between 5 and 50 milliseconds (find further info about image acquisition speed at this link). Consider detecting defects of 100 µm on a steel sheet moving at 120 m/min; in 50 ms, the surface will have moved a total of 100 mm. Since this movement is three orders of magnitude greater than the defect size to be detected, this image acquisition technology is completely unsuitable. Attempting to reduce exposure time by adding additional lighting to the system could potentially lower exposure times significantly, but the associated cost of illuminating sufficiently to achieve this reduction would render the project unfeasible due to increased technical complexity and cost.

This is why optical sensor manufacturers develop linear sensors. Cameras using this technology also produce a matrix image as a result, but it is captured in a completely different manner. While matrix sensors are composed of a total, for example, of 2464×2056 pixels, a linear sensor will be composed of 2048×1 pixels. This linear sensor captures each row of the image individually as the object or surface to be inspected moves, and the camera combines all these rows to produce the complete image of the object.

How does Specular Vision use these Systems?

ISR Specular Vision has differentiated itself from its competitors through the industrialization of cutting-edge inspection technology in complex industrial situations. Therefore, within our portfolio of success stories, projects associated with both matrix and linear technology can be highlighted. In a previous blog entry, we described a successful case in the automotive sector using matrix technology, and in the coming weeks, we will publish another success story in a different sector where linear sensors have been employed. Different images considering automotive aesthetic defects detection can be seen.

Considering systems using linear cameras, it could be highlighted as an example the OIT Residual Oxide Detection system used in the metallurgical industry. This system is intended to inspect stainless steel surface looking for defects from 100 microns (same thickness as a human hair!) to larger visible defects. Images included below are taken using linear technology. As a reminder, the inspected surface is moving while a linear camera acquires a number of rows and puts all of them together in order to compose these images we can see below. There is no difference considering an image acquired using a matrix camera, but industrialization possibilities of linear cameras offer a wider range of possibilities for Optical Inspection Technology (OIT) systems.

In conclusion, the following points can be highlighted:

  1. The foundation of computer vision lies in image acquisition and the industrialization of inspection technology, making the selection of the most suitable image capture sensor necessary for each specific market scenario.

 

  1. Cameras integrating both technologies produce matrix images, although each operates with different timing mechanisms.

 

  1. Matrix cameras capture the entire image at once, whereas linear cameras capture individual rows of the image and combine them to form the final image.

 

  1. Matrix cameras are more suitable for production processes involving discrete parts, while linear cameras adapt better to continuous material production processes. Nevertheless, each case must be carefully evaluated to achieve a comprehensive technical solution for the inspection system.
Comparte la entrada: