Computer Vision Project Restart

Before Covid, I was working on a low-cost computer vision project intended to view a workspace and measure and count material as it is stacked.  With lockdowns and the chip shortage, the project went on hold.

The system is based on the Raspberry Pi, which has been in short supply.  However, the Raspberry Pi foundation has announced that production should be back to pre-Covid levels in 2nd quarter of 2023.

Time to dust off the project.

Depth Cameras

In addition to the Covid-related issues, in 2021 Intel announced they are winding down their RealSense product line.  The computer vision system was initially designed around the D415 depth camera (right).  A lot of effort was expended getting RealSense working with the Rasberry PI.

However, checking Intel’s website, not only is the D415 still available, several other cameras have been introduced.  That doesn’t seem like winding down.  It would be nice to have clear guidance from Intel.

In the meantime, another depth camera was launched through Luxonis, the Oak D, and the Oak D Lite (left).  At $149, the price is a bit more attractive than the D415’s $272.  The specs aren’t as good as the D415, but will it be good enough?  The published depth measurement error rate of 1%-2% at 2-3 meters suggests that it is.

The documentation looks pretty good and there are many published examples of it working with the Raspberry Pi.  Only a prototype will tell.

Wide Angle RGB Camera

A challenge with this project is that the workspace is quite wide at 16 plus feet (~5 meters).  With a typical camera or the Oak D camera, the camera would need to be mounted veryhigh above the workspace, reducing the resolution and increasing the depth error.

The original prototype used a Raspberry PI camera with an external wide-angle lens to reduce the camera height.  The depth camera measures the distance to the workpiece and the RGB camera measures the width and length.

Last week, the Raspberry PI foundation launched a new low-cost $35 12MP wide-angle camera.  Based on a horizontal field of view of 102 degrees the camera could be mounted about 11 feet (3.4 meters) above the deck, well within an acceptable error range for the Oak D depth camera. 

Again, back to a prototype.

Share the Post:

Related Posts