NautilusLanding

Case Study of an IoT Depth Camera

Nautilus 

Following Aquifi's pivot away from gesture-based interfaces, I was tasked with converting the stereo depth sensor we had developed for a set-top box into a ruggedized inspection camera for machine learning-based object identification and defect detection.

These cameras shrugged off the frigid cold of winters in Detroit and the blistering heat of summers in São Paulo, which is more than can be said for those of us who had to install and service them.

During the development of this camera, I owned industrial and mechanical design.

 
 

As any stereo camera system relies on a fixed baseline to maintain its calibration, I designed and ordered these enclosures before the first PCBs shipped from Taiwan.

 

This enclosure was robust, cheap, and well suited to work around the lab.

 
 

Frustratingly, the world was a more dangerous place for our cameras than my lab, so I built a modular test enclosure that would allow us to evaluate new sensor packages without the previous generation's moisture, dust, and RF sensitivities.

 

These were widely deployed, and their internal expansion slots allowed us to rapidly iterate the design of the sensor and its onboard processing.

 

The fact that any externally accessible fasteners could be removed and replaced with a standard camera mounting plate was probably my favorite feature of this generation.

 
 

After mediating a series of negotiations between the software team and the folks who set the target BOM, we locked the optics and the rest of the system.

 

With the help of the folks at Surfaceink, we shrank the system and began preparing for certification.

 
 
 

An eternity of nerve-racking testing later, our Nautilus cameras demonstrated they had met their target of IP67, and we began to build them in earnest. 

 

They are currently used in several applications, most notably in the Bireme dimensioning workstation. 

tempImagepi7TsC.gif