7 Machine Learning Business Ideas

Hi, this post is some kind of giveaway. I want to share a couple of machine learning business ideas I was thinking about. We all have only 24 hours per day so it’s impossible to try out all ideas which come to our minds. Because of that, it’s sometimes worth to share them, i hope they will inspire you. Enough of introduction, let’s get to the list:

  1. Service which will use Machine Learning to allow video streaming without a cameraman, using many cameras, always choosing the one with the most interesting content. Let’s imagine a local football game, currently, you cannot watch local team games if they are not in the top leagues. This service could allow us to easily connect for example 4 cameras and start streaming the game. The app will automatically track the ball, switching from one camera to another, to always present the game from the most interesting angle.
  2. Real-time world mapping for blind people. An app which will use Computer Vision to process image from phone camera (it can be put into a pocket on the chest) and tell a blind person what object is in front of her. Let’s imagine an app that will inform: “you are approaching the pedestrian crossing” or “a rectangular object in front of you” it would be very helpful I believe.
  3. Online courses with Machine Learning support. Currently, if you take an online course to learn to play the guitar or speak in a foreign language it’s the human expert who listens to you and tells what’s wrong and what to correct. With the power of Machine Learning, those people could be replaced with software which will listen, compare with expected result and inform course participant what to change to achieve a better result.
  4. Code supervisor, an app, which will use Machine Learning to analyze written code and suggest how to improve it. In many cases it turns out that we create too big classes, the logic is too complicated, data too public and… no one has time to do a code review. An app that would analyze code in seconds would save a lot of time many software developers.
  5. The Alarm system, which will use Machine Learning and image from home monitoring cameras to recognize dangerous situations, like fire, or leakage before it will cause significant damage.
  6. Price helper, for a lot of used items it’s really hard to estimate what price to ask for when we want to sell them. This app would be able to ask for a couple of details, like item condition, age, color,  and use historical data from eBay/craigslist to estimate how much this item could be worth.
  7. Smart blur, an app that will use Machine Learning to automatically blur some type of objects on video, for example, faces or number plates. 

Ok, that’s it, I shared all Machine Learning related ideas I kept in my Evernote, feel free to turn them into live 😉

In:

7 responses to “7 Machine Learning Business Ideas”

  1. Sergey Avatar
    Sergey

    Hi, dear author.
    Thanks you for your article.
    What you think about idea – create iOS App which use Core ML for recognition of road signs with AR elements?
    I have car and last time think about how use my iPhone as HUD with augmented reality objects? For example, find speed limit sign & display different between it and current car speed.

    1. Tom Baranowicz Avatar
      Tom Baranowicz

      That sounds really cool, if you are interested in this kind of stuff definitely read about openpilot app -> https://comma.ai/

      1. Sergey Avatar
        Sergey

        I know this 2: https://wayray.com/navion & https://www.phiar.net. Last App for me is really cool.
        Now my idea is a hobby.
        I have some problem with training data (photos).
        For example, I use Xiaomi car DVR. It save video files 1 min duration to flash card. I copied this files to Mac, open & take screenshots with any road signs. But I don’t know:
        1. what size should the picture be (1024×768 or 396×396)?
        2. is it necessary to mark only one sign or all that is on the picture?

        1. Tom Baranowicz Avatar
          Tom Baranowicz

          1. It depends on input size of your Neural Network. If you want to train it from scratch (better consider retraining) i think that 396×396 is totally enough, for example yolov3 has input size 416 x 416 and works great.
          2. It is definitely worth to mark all objects you want to discover on each frame/picture.

  2. Sergey Avatar
    Sergey

    IBM Cloud write: “The model resizes the image to 300×300 pixels, so keep that in mind when training the model with images where one dimension is much longer than the other.” I try to do next:
    1. create screenshot from DVR video
    2. crop image with needed road signs as square (square size depends on image min side size)
    What you think? It’s correct?
    P.S: I add email but don’t receive any invite from your answer for example…

    1. Tom Baranowicz Avatar
      Tom Baranowicz

      Hey, your flow looks correct, let me know what results did you get 🙂

      1. Sergey Avatar
        Sergey

        I want to continue the correspondence directly by Skype, Vider, e-mail or any other way…

Leave a Reply

Your email address will not be published. Required fields are marked *