motion recognition with labview

[kupikupi]

hi ... i was wondering if anyone can give me ideas on how to do motion recognition ??
i am working on a project creating a sign language translator. i've already construct a system which recognizes still images. but im now having problems with motion recognition .. my idea is getting the position of the 1st frame and the last frame of a motion as most of the signs has unique 1st and last frame. but i dont think it would work perfectly. is there anyone who could give me some idea?? i also have to do recognition for sentences  where i recognize word by word non stop . so i have to detect which few frames are a word... could anyone hel me .. or push me in the right direction ?? thanx alot...  

[Torba]

Hello kupikupi,

Thank you for contacting National Instruments.  The easiest way to detect motion is to subtract the two images.  This will return the portion of the second image that is different from the first, and from here you can determine what type of motion has occurred.  If you plan to write this code using Vision Builder for Automated Inspection (VBAI), you can refer to the following example on subtracting two images with VBAI:

Performing an Image Subtraction in Vision Builder for Automated Inspection

http://sine.ni.com/apps/we/niepd_web_display.display_epd4?p_guid=04AF5163F3452FADE0440003BA7CCD71&p_...

In regards to text recognition, there is an awesome feature in VBAI that allows you to read text.  This function is called "Read Text", and it is part of the Identify Parts set of inspection steps.  This utility will allow you to "read" text that is found within a search region.  You will first have to train the utility with a known set of characters that you expect it to detect.  Once the utility is properly trained, it will make comparisons between the characters that it sees for each successive image with the defined character set that you have supplied.  For more information on the use of this feature, please refer to the Online Help for VBAI.

Regards,

Mike T

National Instruments

[kupikupi]

hi . thanks for the reply, but actually im not trying to detect motion, im trying to recognize motion by comparing it with the database, i just dont have any idea on what algorithm to use to recognize motion, i have gone thru books and they gave ideas such as subtractiing images, but i dont htink it would work for my case. my idea is that i take the 1st and last image of the motion and compare it with a database but it doesnt work it as the pattern of the motion is similar, but the position will be very far off .. im wondering if there is a way to track the pattern of motion and compare it with the database. i hope somebody can help me, i dont need total help, i just need someone to at least push me in the right direction. thanks.... 

[Torba]

Hey kupikupi,

I can now see where you are trying to go with this, and a great deal of what will need to be done will depend on how flexible you are going to be with the glove user.  Ideally, it would be best to have the user speak with his/her hands in such a fashion that the points of interest (fingers) will always follow defined paths so that it will be easy to compare the profile to a database of defined moves. 

In order to develop a strategy for programming this translator, first define what constraints you must have to the glove user.  For example, lets say that one of the words requires the index and middle fingers on the glove hand to touch the palm of the other hand.  A number of things that will be variable in this move are the rotation of his/her wrist, the starting point of the move, the ending point of the move (where the other hand is positioned), what angle the fingers traveled, and where all of this motion is taking place relative to the position of the camera.  Allowing for more flexibility in all of these variables will greatly increase the complexity in your algorithm, and thus reduce the resolution of your defined moves in the database.  Are there any of these types of variables that you are going to have constrained to help simplify this algorithm?  Not only will this help with code complexity, but it will also reduce the possibility of having two similar move profiles overlap each other in the database causing improperly translated words. 

Regards,

Mike T

National Instruments

[kupikupi]

hi .. thanx for the reply ....

for my project , the things that u mentioned such as the startingpoint, end point , and the position relatice to the camera all cannot be fixed as different gesture can't have the same starting point, it's almost impossible to set a constraint. the only way on my mind now is the check the trajectory of the motion. i just came across some VIs under NI-motion which requires information obtained from vision builder. is it easier if i use vision builder??

i was thinking of a way where i obtain a set of images which represents a word. get the centroid of the glove and plot a graph from there.. same signs by diferent people should have a similar but unique shape. but i dont know how to continue frm there. is there any way to compare similarities of the graph ?? is there any way to improve my concept?? i've differnt approaches but none have worked so far. i have done it successfully for still images where i just get the centroid of the finger tips and compare it with a database with a tolerance of +-10. i can't implement it for motion as it invloves a set of frames.. i hope u can enlighten me.. thanx..

[KevinC]

kupikupi,

Could you post a series of images as an AVI file?

Thanks,
Kevin C.
NI Vision

[kupikupi]

im sorry kevin but i dont really get wwhat u mean ?? do i post a set of images?? or a clip ??thanx

[KevinC]

Any set of images that is representative of what you are trying to process would be helpful. It will make it much easier to come up with some suggestions about your problem.

Kevin C.

[kupikupi]

 hi ... thx for the reply ... im including a set of images which  i have to process.... those images forms a word... i really can't get a any algorithm working.. oi hope  tat u can help me .. thanks..

[kupikupi]

can't open that one.. i'll reattach