About UsBlog
DocumentationGuides, specifications and datasheets
GitHubSource code of our libraries
About UsBlogStore
Jun 27, 2022

Getting Your First Win

Two young engineers get started with OAK
Machine Learning
Computer Vision
One of the biggest challenges with anything technical is the amount of time it takes for your first win. Often, it seems so much of your time is in vain and you spend more time wondering “Can I really do this?” or “Will this ever work?” than basking in the glory of your accomplishment. At Luxonis our goal is to work to minimize this time between you having an idea and seeing it actually working. With that, we'd like to share the story below:
After installing DepthAI (on a Macbook, but every OS is supported) and running the default DepthAI example program, Cole and Noah immediately knew what they wanted to build with their OAK device… A Super Smart Super Soaker But not just any Super Soaker, a Super Soaker that can win any fight automatically; one that can be taught who to go after, for example, an unsuspecting brother. We suspect that Mark Rober might have something to do with this idea being in their heads, but we can’t say for certain.
To put the scope of this project into perspective, a decade ago, a team of some of the most talented electrical engineers, computer scientists, and mechanical engineers all had the same idea for their senior design project to show the culmination of their respective engineering degrees. Some 99%+ of their work (it took over four months!) went into computer vision, and in the end, in order for the soaker to learn where to point, the subject had to wear a specific solid color shirt for the computer to ‘see’ it. It was nonetheless super cool to see and fun to play with. There were limitations, such as if lighting or background changed dramatically, but when it all came together…SPLASH! It worked wonderfully!

Early Wins

Luckily for Noah and Cole, it didn’t take them four months to see success. Within 30 seconds of installing the DepthAI software for OAK-D-Lite, they had: face tracking and identification working by running this model; 3D location of each face working with this model; and semantic depth segmentation of the person using this model. With the three models working right away, the interest and encouragement levels increased greatly, which is extremely important for learning as it lets young engineers know they can accomplish their goals. As the saying goes, nothing breeds success like success. Instead of fighting software, or wondering how to get something working, the near-instant success gave them the brainspace to begin thinking of the next problem to be solved. That next problem was: now that the body itself had been identified, could they tell where specific parts of the body were? With a simple copy and paste of some excellent pose and hand tracking code from geaxgx, the answer was quickly, yes!
Once again, this quick success sparked yet another feature from Noah for the Super Smart Super Soaker project: "Since it knows where my hands are, can I use them to point the super soaker remotely?”  Why, yes you can Noah, yes you can…

Best Laid Plans

Cole and Noah are moving full-force ahead. They have sourced a Pan-Tilt system for holding the squirt gun and also the Pi HAT they will mount to the Raspberry Pi they already have (OAK cameras work great on the Pi series, by the way). It won’t be too long before the plan comes together, the Super Smart Super Soaker will be positioned in some strategic location, and with a wave of a single AI-tracked finger, an unsuspecting family member will suddenly wish they were wearing a raincoat. Good luck on your mission, Cole and Noah.

Brandon Gilles
Brandon GillesCEO