Boost Fertility New Invention Ups Success Rates.pdf
Slide Deck
1. Pooja Venkatesh
MS Research Scholar
IIIT-B
Email: pooja.venkatesh@iiitb.org
Automatic Bharatnatyam Dance Posture Recognition
and Expertise Prediction using Depth Cameras.
SAI Intelligent Systems Conference 2016
21-22 September 2016 | London UK
Dinesh Babu Jayagopi
Assistant Professor
IIIT-B
Email: jdinesh@iiitb.ac.in
2. Index Agenda
Conclusion
Pose Recognition
Expertise Prediction
Geometry of the Bhangas
Ground Truth
Project Flow
Pose Recognition Results
Introduction
3. Agenda
Conclusion
Perform pose recognition for 22 basic postures which form the foundation of
Bharatnatyam.
Finding the origin of these postures based on the geometry of the 4 basic forms called
the Bhangas.
Perform expertise prediction of the dancers based on the recognition results.
Pose recognition and expertise prediction was performed successfully giving an accuracy of
87.14% and 80.80% respectively.
Pose
recognition
Origin
predictions
Expertise
prediction
4. Introduction
Image Courtesy: Ananya, http://www.sehernow.in/ananya2013htmls/kiran-sandhya.html
Bharatnayam is an ancient Indian Classical Dance form originating from the temples of
South India.
“Natyashastra” which is one of the only source of documents for this dance form was
written by a sage named ‘Bharata’ sometime between 200BC and 200AD.
5. Pose Recognition - Selection of Poses
The Bhangas are the basic forms of Bharatnatyam acting as the foundation for different poses in
this dance form.
6. Geometry of the Bhangas
The basic theory of Bharatnatyam assumes the
entire body to be a mass which is equally
divided along an imaginary line that passes
through the centre of the body.
The 22 postures are derived from the 4
Bhangas.
These 22 postures can either be a pure
Bhanga or a mixture of 2 or more Bhangas.
7. Age Group: 13-18 Yrs
Total Dancers: 25
Female Dancers: 23
Male Dancers: 2
Age Group: 6-8 Yrs
Total Dancer: 4
Age Group: 25-35 Yrs
Total Dancers: 15
60
2
Excellent
18-25 Yrs;
37 Female
Dancers
Satisfactory
09-12 Yrs;
17 Female
Dancers
Good
Age Group: >35 Yrs
Total Dancers: 3
Female Dancers: 2
Male Dancers: 1
19
2
Poor
4
15
Ground Truth Statistics
12. The confusion matrix for 22 poses showed similar patterns which could
be grouped to find the parent Bhanga for a set of poses.
A four class classification problem based on the above grouping gave
an accuracy of 93.48%.
This grouping was then used to represent the origin of each of the
poses from their respective Bhangas and verified using Hamming
Distance.
Classification - With grouping
13. 62
21
15
4
NUMBER OF DANCERS
Excellent
Satisfactory
Good
Poor
Expertise Prediction
A difference of joint angles between Excellent
and Poor rated dancers are used as an additional
feature for expertise prediction.
Logistic Regression was used for classification
holding 1/3rd of the data out for testing.
15. A novel problem of posture recognition and expertise prediction for one of the most complex dance
forms has been successfully conducted.
We have estimated the origin of the 22 poses from the 4 basic Bhangas with an accuracy of
93.48 %.
Expertise of a dancer based on age group was predicted with an accuracy of
68.46 % without grouping of labels and an accuracy of 80.80 % with grouping.
Expression recognition and subsequent expertise prediction is being carried out at present.
Conclusion