Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

DIY Object Recognition

244 views

Published on

From my talk at the JavaScript & IoT meetup in Portland, Oregon on August 22, 2017. I talk about using a Raspberry Pi, Node.js, and IBM's Watson Visual Recognition Service to create my own custom classifier.

  • Be the first to comment

  • Be the first to like this

DIY Object Recognition

  1. 1. DIY OBJECT RECOGNITION A TASTE OF
  2. 2. RECOGNIZE STUFF USE JAVASCRIPT TO
  3. 3. INSERT DEMO HERE
  4. 4. DIY OBJECT RECOGNITION TECH STACK ◘ Raspberry Pi Zero W on Raspbian ◘ OV5647-based Pi camera module (v1) ◘ Node.js 8.4.x ◘ IBM's Watson Visual Recognition Service
  5. 5. DIY OBJECT RECOGNITION HOW DOES IT WORK? ◘ It's just an API ◘ Newer Pi's have a camera connector ◘ Raspbian ships with software to interface with camera ◘ Node.js modules interface with software & Watson because JavaScript
  6. 6. DIY OBJECT RECOGNITION CLASS ◘ A set of images ◘ Can be broad ◘ Animals ◘ People ◘ Automobiles ◘ Can be specific ◘ Goats, dogs ◘ Children, adults ◘ Trucks, cars
  7. 7. DIY OBJECT RECOGNITION CLASSIFIER ◘ A logical group of classes ◘ Can contain a class of negative examples ◘ For a classifier containing classes of different animal species, could be plants, people, stuffed animals, etc. ◘ Mostly organizational; improves efficiency when classifying images
  8. 8. DIY OBJECT RECOGNITION CLASSIFICATION ◘ The result is confidence (not "accuracy") ◘ Threshold can be adjusted ◘ Add more images to get more confident! ◘ Multiple classes (and negative examples) may be retrained at once
  9. 9. DIY OBJECT RECOGNITION CHALLENGES ◘ LEGOs are too small! ◘ Resolution? Lighting? ◘ Raspi Zero ain't fast ◘ Training algorithms ain't fast either ◘ Tedious, despite my best efforts!
  10. 10. DIY OBJECT RECOGNITION TRAINING IS TEDIOUS
  11. 11. QUICK TO LEARN, YET SLOW TO MASTER a wiseacre DIY OBJECT RECOGNITION
  12. 12. DIY OBJECT RECOGNITION CHRISTOPHER HILLER ◘ Developer Advocate @ IBM in IoT & Emerging Technologies ◘ Maintains Mocha, a widely-used testing tool for JavaScript ◘ Authored one of the first books on AngularJS, a JavaScript framework ◘ Sometimes cannot put things back together which he has taken apart ◘ Lives in a lovely suburb of Portland, Oregon ◘ Email: boneskull@boneskull.com
 GitHub: https://github.com/boneskull
 Twitter: https://twitter.com/b0neskull
 Blog: https://boneskull.com

×