Object classification use cases are being deployed more frequently, mainly in an industrial context. With this demo, we demonstrate how our Blox platform can be easily integrated in an existing infrastructure, more specifically for doing object classification on a conveyer belt.
- Processor : Nvidia Xavier 16Gb processor integrated in a touchscreen version of our Blox platform
- Camera : The Blox platform is equipped with a 5 Mp embedded camera module from Allied Vision (Alvium 1500 C-500) in order to have a crystal clear view on the chocolates
- The conveyer belt step motor automatically stops/starts when a chocolate is detected. This automation happens via the digital I/O integrated on the Blox platform
The AI model
First, we use a reference to white-balance the image, which helps to automatically adapt the camera parameters to optimize the image quality and to normalize the images. This means that we can capture high-quality images of each praline, which is essential for accurate analysis.
Next, we use foreground-background segmentation, where contour areas are used as robust filters. This technique allows us to isolate the pralines from the background, making it easier to analyze each one individually. This process ensures that we're able to accurately get features for each praline.
Once we have our foreground segments, we use them to compute 17 hand-crafted features. These features capture important aspects of each praline's appearance, such as its shape, color, and texture. By analyzing these features, we can get a comprehensive understanding of each praline's appearance.
Finally, we use machine-learning techniques, specifically decision trees, to classify the selected pralines into their respective categories. This means that we can quickly and accurately label the pralines according to their characteristics and display the contents of each praline.
This results in a highly-qualitative demo, and working prototype, created in less then 2 weeks (from scratch), showcasing the flexibility of our Blox platform to easily demonstrate the added value of your edge AI initiatives.