Machine Vision - Powering Self-Driving Cars and More 

At SHAKO, technological innovation is at the heart of what we do.  Machine vision is an essential tool for allowing machines to interact with their environment. Through cameras, sensors, and algorithms, we're now decoding real-time visuals, heralding advancements in areas such as facial recognition, self-driving vehicles, and augmented reality. Read on as we explore the technical layers of machine vision, and discover how SHAKO is geared to bring this expertise to your next venture.

Machine vision systems start with capturing an image, predominantly through visible light cameras but sometimes with sensors such as lidar or radar. The camera, lens, and lighting needs to be designed specifically for the application in order to get the highest quality sharpness, exposure balance, and clarity of the critical areas of the image. Often machine vision systems will capture multiple images, each with different lighting or lens configurations, in order to highlight certain aspects of the scene. These images become the raw data for the downstream algorithm, see an example below. 

The image on the left is shot with natural lighting. The image in the center is backlit and highlights the outer edges of the widget, but loses some of the edges and texture within the part. The widget can also be emphasized using a Sobel filter (right), which would take the naturally lit image and mathematically detect the sharp transitions between light and dark, highlighting all the edges and texture.

With the images acquired and filtered, the system then identifies specific features, including shapes, dimensions, and colors. A common technique is to compare the shapes in the image to a known ideal shape. This is especially crucial in settings like manufacturing, where precise component placement or defect detection is vital. The system can detect minor deviations from the expected shape, often times not detectable by the human eye.

Imagine a conveyor belt in a manufacturing setup, streaming automotive components. Machine vision cameras can detect imperfections in fractions of a second and trigger a mechanism to sort the acceptable parts.

Another technique used for shape recognition, where a single training image does not adequately describe the object, is neural network shape recognition. Imagine that for a self-driving car you want to detect a pedestrian in a street intersection - but there are endless ways in which a pedestrian might appear. A neural network can be used to detect shapes that are likely to be a pedestrian, based on what pedestrians typically look like. Neural networks are inspired by our brain's architecture, consisting of interconnected associations or "neurons". Through training on vast datasets, these systems learn to identify and recognize various shapes, making decisions based on the patterns they've been exposed to. You can leverage pretrained models, or train your own to become adept at distinguishing even the most complex objects, making neural networks invaluable in diverse applications like self-driving car algorithms. 

As an example that utilizes both techniques, think about smartphone photography. Edge detection helps adjust the camera's focus, while shape recognition via neural networks optimizes subject highlighting, or adds a dynamic background behind the subject.

In essence, machine vision, with tools like shape recognition, is charting the future of precision and interactive technology. If you're eager to leverage this power or have a project in mind, remember SHAKO is ready to collaborate. Reach out to SHAKO today and together, let's elevate your business.

If you enjoyed this content, hit the subscribe button below so you never miss a post. It's free!

Please help us out by sharing this article with friends and colleagues!

About the author: Justin Pratt