Haarcascade or HOG for lego identification ?
Hi everyone,
I would like to devellop a system to identify lego blocks, the aim is to verify if a lego set (a set of different lego blocks) is full using a single image for identification.
The system must be able to identify a block from different views.
Do you think the Haarcascad method or the Latent SVM method could be used for this problem ? (it seems that Haarcascade must be already excluded, because it's not Rotation invariant).
Thank you for your answers and have a good day !
Gausspell
Since you know exactly what to expect, I would not look at functions that generalise over appearence like DPM, SVM, Cascade Classifiers, ... instead look at bag of visual words, keypoint matching, ... which are far more robust to viewpoint changes!
As to excluding haarcascade because it is not rotation invariant, that is bogus. There are possibilities to combine different views in different classifiers and then model them as a forest of cascade classifiers.