Research Experience

August 2008 –  Present    Graduate Research Assistant, University of Central Florida, USA
Advisors: Prof. Mubarak Shah, Dr. Pingkun Yan

Video Image Retrieval and Analysis Tool (VIRAT) project (Aug 2008 – Feb 2012)

  • Formulated experiments to test different action recognition algorithms on APHill aerial videos provided by DARPA which were recorded using Electro-Optical/Infra-Red (EO/IR) sensor from a military aircraft flying at a height of over 1000 meters. The videos are aligned and the moving objects are tracked using in-house system COCOA.
  • Integrated two action recognition algorithms “Bag of Visual Words” and “Lagrangian Particle Trajectories” into the VIRAT DARPA system developed by Lockheed Martin and Kitware as part of VIRAT project. The algorithms were converted to C++ for integration.

Action recognition (Jan 2009 – present)

  • Developed an algorithm to perform multiple action recognition, incremental action recognition and action localization using feature-trees. ( Jan 2009 – Jun 2009)
  • Demonstrated an action recognition system “KEYS” live at ICCV2009 conference held in Japan. It is developed in C++ based on feature-trees algorithm for real-time action recognition. New action categories can be added to the system on the fly. (Mar 2009 – Sep 2009)
  • Demonstrated improvement in accuracy on different action recognition datasets by using scene context information in conjunction to any motion descriptor. The datasets include UCF11, UCF50, HMDB51 and a sample from TRECVID dataset released as part of ALADDIN project. (Oct 2010 – May 2011)
  • Developed a new video representation “Particle Flow Fields” based on optical flow and particle advection and experimentally showed that it outperforms optical flow and gradients. Particle Flow Fields showed good performance on Kinect Gesture Recognition dataset in a “One-Shot-Learning” framework. (Sep 2011 – present)

GPS location from image/video (Jun 2009 – Jul 2009)

  • Involved in developing a proposal that uses the concept of feature-trees and local interest points to geolocate images and videos. The project was eventually funded by Intelligence Advanced Research Projects Activity (IARPA).

Smartphone to detect human actions (Jan 2011 – May 2011)

  • Conceived and helped solve the task of performing action recognition using data from accelerometers on smart phones. UCF-iPhone dataset was generated by recording data from an Inertial Measurement Unit (IMU) on an Apple iPhone 4 for 9 different actions performed by different actors.

Image classification and segmentation (Jul 2009 – Oct 2009)

  • Developed a framework to use feature-trees and modified Bag of Features approach for image classification and segmentation in PASCAL dataset.

Brain tumor detection and segmentation in MRI images (May 2009 – Nov 2011)

  • Developed a discriminative approach for automatic brain tumor detection and segmentation in MRI images for a project funded by National Institutes of Health.
  • Demonstrated the use of multiple modalities in MRI and also experimentally showed the use of texture features like LBP and HOG along with intensity to train a discriminative classifier to do reliable pixel level tumor detection.
  • Also developed a framework to use confidence scores from discriminative classifiers to improve the accuracy in different segmentation algorithms.

Datasets released for public (Jun 2010 – May 2011)

  • Led a team of 10 people to develop one of the biggest action recognition datasets: UCF50. This dataset has about 6676 videos taken from YouTube and has 50 different action categories. (Jul 2010 – Oct 2010)
  • Also, led a team of 15 people to develop a unique action recognition dataset UCF-ARG where the action is simultaneously captured from a ground camera, a rooftop camera at 100 feet and a remotely controlled aerial camera mounted on a helium balloon. (Jun 2010 – May2011)

May 2011 –  August 2011    Research Assistant (Intern), Kitware Inc., Clifton Park, NY, USA
Advisors: Dr. Naresh Cuntoor, Dr. Amitha Perera

  • Investigated the performance of 3D gradients for action recognition task in aerial videos taken from an aircraft and performed extensive testing on aerial and rooftop videos in VIRAT dataset. The designed 3D-STHOG descriptor is incorporated into VIRAT DARPA system for testing.
  • Analyzed the sensitivity of 3D-STHOG to scale, frame rate and also to translation of bounding boxes that might occur due to errors in tracking and tested for performance on six different action recognition datasets.

August 2006 –  August 2008    Graduate Research Assistant, University of Central Florida, USA
Advisor: Dr. Aman Behal

  • CONTROL DESIGN: Designed adaptive “Multi Input Multi Output” (MIMO) control for aeroelastic vibration suppression using Tuning Functions, Adaptive Back Stepping and Output Feedback methods.
  • SYSTEM IDENTIFICATION: Employed Volterra kernels for non-parametric modeling and system identification on systems like “Gammamotor Neurons”, “Aeroelastic Wing Section”, and “Circulating Fluidized Bed Systems”.

January 2006 – August 2006    Graduate Research Assistant, Clarkson University, USA
Advisor: Dr. Aman Behal

  • Designed an “Adaptive Output Feedback” controller for aeroelastic vibration suppression on a nonlinear wing section.

July 2005 –   January 2006    Research Assistant, Haute école valaisanne, Sion, Switzerland
Advisor: Dr. Gilbert Maitre

  • JPEG2000 technology was tested using Kakadu Software and re-designed to be implemented on an Embedded System to investigate the advantages of Co-design.

August 2003 – July 2005    Research Assistant (Volunteer), Fachhochschule Südwestfalen, Germany
Advisor: Dr. Gerhard Petuelli

  • Signal Processing for Predictive Maintenance: Data Acquisition was done on CNC machine using vibration sensors and data was analyzed to investigate faults in the machinery and also predict break downs using neural networks and fuzzy logic.

January 2002 – April 2002    Internship, Indian Space Research Organization, Sriharikota, India
Advisor: Mr. K. Srinivasa Rao, Prof. P. Nageswara Rao

  • “Microcontroller   Based   PCM Frame Synchronizer” was built on PC ADD-ON CARD compatible to an ISA Bus (Project done as a part of my final year B.Tech).