Usually my posts are finite…I get my idea on digital paper and put it out there for the world to see, but I wanted to do something a little different.  The idea here is to have this post be your glimpse into a project I have in mind.  A project where I can get more hands on developing my own Artificial Intelligence using computer vision, a type of machine learning where you train a machine to recognize objects or actions.  My intention is to continuously update this post with developments on the project and hopefully some positive outcomes.
This is serious business…

Background:

  • I have a yard and I like it a lot.  I like to walk barefoot in it and so does my family.
  • People often let their dogs poop in it, but break city law and leave it in my yard for me to enjoy (I don’t enjoy the feeling of any fecal matter between my toes or the process of removing said fecal matter from said toes or yard).
  • Social media is a thing now, believe it or not.  Even though I rarely talk to my neighbors in any real human interaction, we are closely connected on social media with Facebook groups and Nextdoor being all the buzz with neighborhood goings on.
  • IP cameras are cheap and I have plenty of them.
  • I am interested in computer vision, machine learning, and artificial intelligence.

Hypothesis:

  • If I (preferably an AI of some sort) begin posting pictures of the people (not naming them) who leave their beloved pups poop in my yard (breaking city law no less) on neighborhood social media, then I will see a marked reduction in forsaken canine fecal deposits in my yard.
  • If successful (or at least capable of grabbing a few pics of poop and purveyors of said poop in the act), I am pretty sure this is both a media-worthy endeavor and probably a pretty fun conference talk.  Would you want to hear this talk?

Process:

  • Research
    • Are there existing models that exist for this particular use case?
    • If adjacent models exists, what is the lift to modify?
    • If I have to build from scratch, what data sets will be most useful for training?
  • Training
  • Testing
  • Retraining
  • Prototype testing with no posts, but alerts for inspection
  • Retraining
  • Implementation
  • Marketing
  • Heavy social shamming of those leaving poop.

Considerations:

  • If its a minor/child is the one leaving poop from their precious pup, I probably don’t want to shame them…same might go with disabled people.
  • If it’s a cat, should I consider an AI powered weapon? (just kidding cat lovers)
  • With the advent of surveillance and doorbell cameras, does anyone care I have a camera in my yard?  Do I care that they care?

Tools to consider:

  • Maybe Tensorflow?
  • Do I need a cluster or will a single robust machine be able to handle this?
  • Can I automate this with AWS Rekognition?
  • Is DeepLens the easy button here?

So here’s the deal…

I am going to begin working on this in my margins in the weeks ahead, but I would welcome any feedback or guidance from experts out there that can help me accelerate my project.  Email me at cory@bigdatabeard.com or feel free to comment below.

I’ll post my progress to this site as it develops.  Again, the goal is to stop the orphaned poop in my yard but I will settle for a fun learning outlet and something maybe worth talking about at an obscure conference next year.

Your bearded friend, Cory