Ask Your Question
0

what kind of approach can be used for recognize hand signs?

asked Jan 22 '14

Parthi gravatar image

updated Jan 23 '14

what approach is used in the below video to recognition for hand signs?

Sign Language recognition

why is he using cvblob? how did he separate his head from hand ( it has different colors) how to actually train the signs using the SVM ? how much time will it consume ?

Edit:My questions

  1. how to recognize signs used in a sign language ? do we need store the signs as files in a Haar-like database? is it possible?
  2. if we skin color recognition is used ? how to identify the signs?
Preview: (hide)

Comments

1

hi, parthi, your youtube link is no great help, most ppl won't even look at it.

chances for feedback / answers are much better, if you write down your problem.

berak gravatar imageberak (Jan 22 '14)edit

2 answers

Sort by » oldest newest most voted
1

answered Jan 22 '14

Vikas gravatar image

Instead of cvblob, you can use findContours in opencv to find blobs of same color.

Use face detection to separate head from hands.

You will have to train one vs all SVM for this multiclass problem.

Preview: (hide)

Comments

1

true, cvblob is totally outdated / overrated nowadays

berak gravatar imageberak (Jan 22 '14)edit

I do not agree on the one vs all SVM approach, you can perfectly run a single SVM with yes or no for each sign, then look at which positive responses show the highest confidence :)

StevenPuttemans gravatar imageStevenPuttemans (Jan 23 '14)edit
1

answered Jan 23 '14

MRDaniel gravatar image

Skin detection to create a mask. Use a classifier for pixels in hand such as EM() to create a cleaner map (optional). Find contour of hand.

1) Compare to known shapes using matchShapes().

2) Create a database of known shapes find some features to classify. Train classifier and then use for comparison on any new shapes found with this method.

Preview: (hide)

Comments

is there a link or a tutorial for step 2 . Exactly how to create that database and train the classifier

Parthi gravatar imageParthi (Jan 23 '14)edit

I suggest you go read some machine learning book about classifiers :) google or search this forum for SVM, cascade classification, Viola & Jones, naive bayes, regression trees, ... all possible approaches for your problem!

StevenPuttemans gravatar imageStevenPuttemans (Jan 23 '14)edit

Question Tools

Stats

Asked: Jan 22 '14

Seen: 1,365 times

Last updated: Jan 22 '14