Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 16

Android Based Face Detection and Tracking (ISpy)

Sanmay Patil Rucha Ingle Sagar Joglekar

3/29/2013

Project Idea
Explore computationally expensive Image processing on Android platform Develop for some real-time application Explore Advanced optimizations Portable Surveillance: a neat Idea.

3/29/2013

OpenCV
JNI enables developers to access C/C++ libraries OpenCV: a fast way into Image processing Needs Cross compilation

3/29/2013

Cross compilation Procedure

OpenCV 1.1 Fragments:


CV CVAUX CXCORE OTHERLIBS ML (Optional)

Makefile to link all the functions in these fragments except those needing GTK, QT and python Support
3/29/2013 4

Cross compilation Procedure


Simply compile using ndk-build (need to solve the dependencies on the way) Easier way: Download from our site Download into the jni folder and you are set.

3/29/2013

Application specs:
App that detects breach in security and gives some sort of indication to the owner Face Detection and Tracking.

3/29/2013

Application structure

3 Intents to Distribute work flow.


Detection and notification Detection and tracking Email Configuration

3/29/2013

Main

Preview Thread

Image Display

JNI
Set Source Progress Get Source

Display

Send Email
3/29/2013 8

Algorithm

Face Detection
Figure 1 . Exampl es of the Haar features used in OpenC V

Fig 2 The Haar featur es on face

3/29/2013

Face Tracking
Celebrated Mean Shift algorithm

Histograms

The face-probability view,

Despite many limitations, due to high computational efficiency good choice for embedded platforms.
3/29/2013 10

Demo

http://www.youtube.com/watch?v=wmf 8J8ItPbs

3/29/2013

11

Optimization
Optimized face detection with increasing scale factor of integral images Didnt use nested faces cascade For tracking, made the searching and localization area smaller No need to process every frame

3/29/2013

12

GPGPU was first choice, however there are very few important functions which are not iterative Out of candidate functions like cvIntegral loading textures and retrieving, voids the advantage of parallel computation Comparison with cascades is viable option OpenGL ES 2.0 has many limitations like limited length of shader programs More suitable for CUDA platform

3/29/2013

13

Future work
In the spring quarter, continuing the project for research credits with the emphasis on GPU/Tegra Platform To implement more accurate and efficient tracking algorithm Market Release

3/29/2013

14

References

http://www.stanford.edu/~zxwang/index.html : Cross compilation help http://www.cognotics.com/opencv/servo_2007_series/index.ht ml: Good Tutorial for face detection and Tracking http://fileadmin.cs.lth.se/graphics/theses/projects/facerecognit ion/thesis.pdf : Face detection on GPU http://research.microsoft.com/enus/um/people/viola/Pubs/Detect/violaJones_CVPR2001.pdf : Viola Jones Paper http://coewww.rutgers.edu/riul/research/papers/pdf/trackmo.p df: Mean Shift Paper http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=5396 139 : Fragment Based Tracking

3/29/2013

15

Thank You

3/29/2013

16

You might also like