0

Currently, I know that Vuforia can track objects that are scanned in the targets database. This means that if I wanted to recognize many objects, I have to scan each of them and upload them into the database.

Here’s what I want to create. I want to create an app using Unity that can recognize and track simple shapes (such as a rectangular box or a spherical ball) disregarding the size. For example, the app should be able to recognize a small Kleenex box as well as a big shipping box.

Is it possible to recognize these shapes with minimal object targets (e.g. recognize all rectangular objects with one object target)? If so, how can I set up vuforia to do so? I understand that this question may be confusing to understand, and I will do my best to elaborate more clearly. Thanks so much!

Jsbbvk
  • 180
  • 2
  • 19
  • Having used Vuforia's object recognition...this is not possible. Vuforia ***hates*** round shapes. Cubes and similiar on the other hand have a few readily identifyable corners, but it leads to very poor tracking quality. Vuforia wants, at a minimum, 300 individual tracking "features" to identify in order to accurately calculate the object's position. Features are anything that a black and white camera can identify as a "corner" (hard color transition and hard angles; less than around 130 degrees). – Draco18s no longer trusts SE Jun 10 '18 at 03:21
  • @Draco18s thanks for your answer! So vuforia can’t recognize a simple box because it has no unique features? The goal of my app is to recognize simple shapes (without scanning every object) and stick a cube on it. So if the user moves the shape, the cube will still be attached to it. Is this idea possible with vuforia? Thanks once again! – Jsbbvk Jun 10 '18 at 03:31
  • It is not possible with Vuforia as far as I am aware. You would need to write your own identification code. – Draco18s no longer trusts SE Jun 10 '18 at 04:21
  • You can't expect a system to recognize something it does not know about. Same when you looking at something you've never seen. You can train your system, but this is off topic for now. What you'd need to do is provide the system with a set of similar shapes that give the same result. A large box, a thin box, a small one, a cubic and so on. All those would give the result of "box". Except if you know about machine learning, this is the only way for you to achieve this. – Everts Jun 10 '18 at 06:53

0 Answers0