How to recognize hand on android devices

edit

I write the follow code according to handDetection.cpp,but the result is negative .What's wrong?By the way do you know which functions in opencv-android library can replace the "cvAnd" and "cvOr" functions?

the code :

Scalar hsv_min = new Scalar(0,30,80,0); Scalar hsv_max = new Scalar(0,30,80,0); Imgproc.cvtColor(src, hsv_image, Imgproc.COLOR_BGR2HSV); Core.inRange(hsv_image, hsv_min, hsv_max, hsv_mask); Utils.matToBitmap(hsv_mask, bmp);
image.setImageBitmap(bmp);

You should first check the color of the hand pixels with given lighting conditions and hsv limits should be set accordingly.

Here is the OpenCV java documentation: http://docs.opencv.org/java/ . The Core class contains bitwise_and and bitwise_or.

Thank you,rics! I already recognized the skin,then i have to distinguish hand from face.Do you have good advice?

You can use contour detection. Hand skin may have concave parts if fingers are open. You may detect face with a Haar cascade classifier and use only negative answers if only face and hands are on the image.

I used convexHull and convexityDefects function to find hand convex hull and the convexity defects of it in these day,but i didn't know what should to do for distinguish them yet? By the way,if i want to read the number that the gesture showed,how to realize? I searched in the internet much ,but nothing helpful. Can you help me ? I am a fresher, really appreciate !

There are convexity defects on an open palm. If you find a skin colored objects with convexity defects it can be a hand. For closed palm you can try another Haar classifier: http://www.andol.info/hci/2059.htm. Moreover a similar question appeared here: http://stackoverflow.com/questions/9168785/hand-detection-using-opencv .

Hi,rics,I have visited your website,saw the MINDdroidCV,but I can't watch the videos, can you show me them? By the way ,could you calculate the distance between the robot from and light blob?

They are on youtube, so you could watch if your computer is properly set. You may try to watch it from here: http://www.youtube.com/user/rics5. I have not tried distance calculation. I would use a known object with distinctive colour and I would count the pixels of the object colour (with zeroth moment) on the image from some measured distance during the learning phase. Later I would count the pixels again and interpolate from the previously learned distance table. I do not think it is a really reliable method.