Object recognition using this code

[maxi-mus]

Any tips on how I could extend this project to recognize the learned image in a live video stream for example?

[mbrenon]

Hi maxi-mus,

This project is mainly a proof-of-concept experiment on the SIFT extraction alone. Implementing full image recognition based on this project would require a lot of work:

• there's some wrapping to do aroung the sift extractor in order to be able to feed it camera frames,
• only the SIFT extraction is handled! This means that you would also have to implement:
  • a "reference" feature vectors database,
  • a search among this database to find and validate a match.

In addition to this, the feature extraction based on this experimental method is not very efficient: you'll get better performance using a third-party library!

[maxi-mus]

Ok, thanks. Any third party libraries in particular you could recommend?

All I'm really trying to do is use a pre-existing library to extract some features, and then run a sliding window on live video (on an iPhone) to detect them. I've looked into a couple of projects (GPUImage, DeepBeliefSDK, this one) but they all won't help me due to various reasons.