ios – CV, ML beginner wants plan of assault for cell app’s sport card detection


I have been a developer for many years however have not labored a lot with laptop imaginative and prescient and/or utilizing machine studying. I’ve a proof-of-concept private undertaking that touches on each, presumably, and I lack sufficient expertise to make affordable selections about proceed. So I am searching for details about the appropriate strategy, and presumably about current software program to make use of to get me to the purpose of constructing working code.

A couple of years in the past a good friend created a set of playing cards for a tabletop sport of their very own devising. They’ve since added different playing cards to the sport. They’ve had average success with it and since I have been constructing iOS apps for the reason that first iPad appeared, they requested if it will be potential to make an app that might develop the sport into the app world. As we mentioned the possibilites the factor that excited us probably the most was the concept of pointing a tool at some gameplay on a desk and have the app determine the playing cards and their association. This could possibly be utilized in quite a lot of methods so as to add new capabilities to the sport which will or might not be helpful or enjoyable.

So my query is, can this be performed? Can we make a toy app that demonstrates the detection and permits us to see whether it is truly value doing? I first set out by utilizing the ARKit in iOS. I shortly bumped into limitations of that library. I’ve an 80 card deck plus comply with on card units, and ARKit simply is not as much as that, I might get to work with a handful of playing cards at most.

I learn articles on laptop imaginative and prescient and that led me to OpenCV. On my laptop computer I created python code that matches any of the 80 full playing cards in a photograph of some arbitrary playing cards organized on a desk. It’s a bit sluggish and as I labored to hurry it up the variety of mistaken matches elevated. Some playing cards share visible parts although every card is distinct to the human eye.

There is likely to be higher options in OpenCV however as I learn up on related issues it appeared that what I would strive subsequent is to create a machine studying mannequin and put that mannequin into my iOS app and have it use the mannequin to do the detection for me. I wish to strive that, however so most of the steps wanted are new and I would prefer to keep away from unhealthy selections and even the entire strategy whether it is incorrect. As an illustration, related issues appear to be being solved by utilizing Keras and TensorFlow, however I do not know if that is proper for my drawback.

And I barking up the incorrect tree?

Can anybody record the appropriate sorts of issues I must be doing if that is the appropriate path?

Simply so as to add element in case it wasn’t clear. I am not fascinated about classifying actual world object sorts, I wish to detect my 80+ card photographs as they could seem on anybody’s desk. The person might need to maneuver the gadget or add mild so the digicam sees them clearly. The cardboard photographs are fastened and the maker provides new card extensions over time. I wish to determine and save every largely seen card and retailer their positions, and possibly the picture. Thanks!


Leave a Reply