Hand-Forearm Segmentation and Gesture Recognition Using Logistic Regression
1. Under The Supervision of:
Dr. Kishor Sarawadekar
By:
Vandit Chauhan (14095076)
Shivam Agarwal (14095063)
Aman Soni (14095004)
2. ABSTRACT:
Hand gesture recognition applications requires a reliable
identification of the hand region and its subdivision into
fingers and palm areas.
The center of the palm and the hand orientation are
identified.
Then circular and elliptical shapes are fitted on the
extracted samples in order to reliably identify the palm and
fingers area.
The proposed approach has been tested on a given dataset
and preliminary results show its reliability.
3. Hand-Forearm Segmentation
Segmentation of the hand from an
image is a necessary step for many
applications. e.g. hand tracking,
gesture recognition. One of the
major problems is the hand-
forearm segmentation. The
experimental results prove that
proposed algorithm is accurate
and fast.
4. Hand-Forearm Segmentation
In the beginning,
distance transform of
the image is obtained,
and then the pixel
having the maximum
value gives us the
center point of the
palm.
5. Hand-Forearm Segmentation
Then the orientation
of the hand is
determined, and in
accordance to that the
arc joining the palm
with forearm is
determined, as
shown.
6. Hand-Forearm Segmentation
The point on the arc
having maximum value
of distance transform,
gives us the wrist point,
and a tangent is drawn
at the circle at this point.
The region below the
tangent is eliminated.
8. Palm-Fingers Segmentation
From the distance
transform obtained
previously, the centroid is
chosen as the centre and
the maximum value is
chosen as radius, for
drawing the circle shown in
the figure.
9. Palm-Fingers Segmentation
Now the circle is traversed
and at each point, the
region is grown by choosing
the point as center and the
distance transform value of
the point as the radius. The
grown region is shown
alongside.
12. Gesture Recognition
How to recognize the gestures ?
Understanding the dataset :
• Divided into various classes based on number of
fingers.
• Further divided into subclasses based on different
types of gestures.
13. Gesture Recognition
Class 1: Gestures with one finger. Contains
3 Subclasses:
Subclass 1:
Subclass 2:
Subclass 3:
14. Gesture Recognition
Class 2: Gestures with two fingers. Contains
3 Subclasses:
Subclass 1:
Subclass 2:
Subclass 3:
15. Gesture Recognition
What next ?
Create classifiers for various subclasses.
Train the classifiers using logistic regression
with the help of data available.
16. Gesture Recognition
What features can be included:
Pixel value of fingertips.
Centroid of palm.
Relative angle b/w fingers.
Centers of fingers.
Orientation of hand.
Angle b/w palm center and fingers.
17. Gesture Recognition
So how do we proceed ?
Create a hypothesis function for all
subclasses.
The hypothesis function returns probability
of an image, to belong to a particular subclass.
18. Gesture Recognition
The logistic regression hypothesis is
defined as:
where function g is the sigmoid function.
The sigmoid function is defined as:
19. Gesture Recognition
The cost function in logistic regression is
given as :
Then by using gradient descent to minimize
the cost function, the various unknowns in
theta matrix is determined. Hence the
classifier is trained.
22. Applications of Hand-Gesture Recognition
There are numerous applications of hand-
gesture recognition. Some of them are
listed below :
• Sign Language for Blind
• Hand-Gesture Controlled Robots
• Virtual-reality Gaming
• Gesture Control for TV
23. Applications of Hand-Gesture Recognition
Gesture Control for TV:
Now-a-days, some smart TVs have the feature
where you can control the TV through hand
gestures such as increasing/decreasing the volume,
changing the channel, etc.
24. Applications of Hand-Gesture Recognition
Gesture-Based Gaming:
Again most of the actions in a game can be
performed using hand-gestures which makes the
game more realistic and adds fun.
25. Applications of Hand-Gesture Recognition
Hand-Gesture Controlled Robots:
Gesture recognition can be used to create a
wireless-controlled robot.
The various classes can be assigned a specific
task, such as locomotion.
The subclasses can be used for performing the
operation, such as moving forward, backward, in
case of locomotion.
26. References
• Zhi-hua Chen, Jung-Tae Kim, Jianning Liang, Jing Zhang, and Yu-Bo Yuan
“Real-Time Hand Gesture Recognition Using Finger Segmentation”
Hindawi Publishing Corporation
The Scientific World Journal
Volume 2014, Article ID 267872, 9 pages
• Bosheng Wang, Jiaqi Xu
“Accurate and fast hand-forearm segmentation algorithm based on
silhouette”
2012 IEEE 2nd International Conference on Cloud Computing and
Intelligence Systems (Volume:02 )
• Giulio Marin, Marco Fraccaro, Mauro Donadeo, Fabio Dominio, Pietro
Zanuttigh
“Palm area detection for reliable hand gesture recognition”
Department of Information Engineering
University of Padova