iccv
tracking
face tracking

Machine Learning on Positive Definite Tensors

Symmetric positive definite tensors in the form of sample covariance matrices are very popular in visual surveillance, DT-MRI imaging, face recognition, etc. Our research aims at developing algorithms for efficient computations on these tensors. 
  1. Dirichlet Process Mixture Models on Symmetric Positive Definite Matrices for Appearance Clustering in Video Surveillance Applications, A. Cherian, V. Morellas, N. Papanikolopoulos, S. Bedros, Computer Vision and Pattern Recognition, 2011
  2. Efficient Similarity Search for Covariance Matrices via the Jensen-Bregman LogDet Divergence, A. Cherian, S. Sra, A. Banerjee, N. Papanikolopoulos, International Conference on Computer Vision, 2011
  3. Generalized Dictionary Learning for Symmetric Positive Definite Matrices with Application to Nearest Neighbor Retrieval, S. Sra and A. Cherian, European Conference on Machine Learning, 2011,  
image database    image search results

Image Similarity Search in Large Visual Datasets

Given a large database of images, and a query image, this project addresses and propose novel algorithms for finding approximate nearest neighbors from the image dataset. Our algorithm is based on the recently introduced paradigms of sparse coding and dictionary learning.
  1. Efficient Similarity Search via Sparse Coding, A. Cherian, Va. Morellas, N. Papanikolopoulos, Transactions on Pattern Analysis and Machine Intelligence (TPAMI) (under review), (UMN Tech report)
  2. Denoising Sparse Noise via Online Dictionary Learning, A. Cherian, S. Sra, N. Papanikolopoulos, Intl. Conf. on Acoustics, Speech and Signal Processing, 2011
  3. Approximate Nearest Neighbors via Dictionary Learning, A. Cherian, V. Morellas, N. Papanikolopoulos, Proceedings of SPIE, Orlando Florida, 2011

video search

Activity Recognition and Video Content Search

This project aims at detecting and recognizing abnormal activities from large video databases. We apply the technique towards medical and behavioral diagnostic applications.
  1. A Multi-Sensor Visual Tracking System for Behavior Monitoring of At-Risk Children, R. Sivalingam, A. Cherian, J. Fasching, N. Walczak, N. Bird, V. Morellas, N. Papanikolopoulos, G. Sapiro, K. Lim, Intl. Conf. Robotics and Automation, 2012

object annotation

Object Recognition and annotation via contextual reasoning

The primary goal of this research is to develop a context-based reasoning system for the task of automatic image annotation. That is, if we assume that a pure feature based object recognition is never going to be accurate enough, can we improve the object recognition process through inferring the context in which the object exists? This has applications in image search engines, video surveillance and scene recognition.
  1. Automatic Image Annotation using Region Covariance
    Descriptors, A. Cherian, (unpublished work, details available on request)

  2. Compact Covariance Descriptors in 3D Point Clouds for Object Recognition, D. Fehr, A. Cherian, R. Sivalingam, S. Nickolay, V. Morellas, N. Papanikolopoulos, , Intl. Conf. on Robotics and Automation, 2012
altitude estimation

Machine Learning and Vision Algorithms for Flying Robots

In this project, we investigate the possibilities of deploying machine learning and computer vision algorithms for the autonomous navigation of Unmanned Aerial Vehicles (UAVs) in indoor environments. Our research differs from similar work in the area in that we try to push the limits of using a single onboard camera towards estimating the speed, altitude, state, etc. of the UAV by combining the vision algorithms with machine learning algorithms.
  1. Motion Estimation of a Miniature Helicopter using a Single Onboard Camera, A. Cherian, J. Andersh, V. Morellas, B. Mettler, N. Papanikolopoulos, American Control Conference, 2010

  2. Autonomous Altitude Estimation of a UAV using a Single Onboard Camera, A. Cherian, J. Andersh, V. Morellas, N. Papanikolopoulos, B. Mettler, Intl. Conf. on Intelligent Robots and Systems, 2009
3D reconstruction

3D Reconstruction

Given a single photograph of a scene, can we reconstruct the 3D environment in the image? Humans do not have much of a difficulty in inferring the three dimensinal world with one eye closed, because we "know" how the world looks like! In this project, we try to put this "knowledge" into the computer through a supervised learning model!
  1. Accurate 3D Ground Plane Estimation from a Single Image, A. Cherian, V. Morellas, N. Papanikolopoulos, Intl. Conf. on Robotics and Automation, 2009
image quality

Evaluating Quality of Images

Computational aesthetics is an up and coming interdisciplinary field that bridges
aspects of philosophy, psychology, art and computer science. The basic goal
of this new field is the analysis of creative behaviors, along with methods to
augment them using computational approaches. In this project, we investigate
aesthetics of images from a statistical perspective using machine learning
techniques. The basic approach of the project is to learn an association between
potential aesthetic qualities of photographs against their ratings by people, later classify a given image as belonging to a professional category or an amatuer one!

  1. Evaluating Aesthetics of Photographs via Supervised Learning, A. Cherian, (unpublished work, details available on request).

Simultaneous Localization and Mapping

Suppose we are sending a robot to Mars. We do not really know the terrain of Mars, nor we have a map of this area. So if the robot is to move around in this area without getting lost, it has to find landmarks, and build a map by itself! This is the classical problem of Simultaneous Localization And Mapping (SLAM). In this project, we implemented SLAM algorithms based on Extended Kalman filters.
  1. Simultaneous Localization and Mapping, CSCI5551 project together with R. Sivalingam and M. Prahladka, supervised by Prof. S. Roumeliotis
    


Automatic Robot Parallel Parking


Parallel parking of a car, or any vehicle for that matter, is the most difficult manuever that a driver has to fiddle with during driving. So what if we impart the knowledge of doing this to the car itself, so that when a button is pressed, it can find by itself the best place to parallel park and do the parking itself? This is what we try to achieve in this project, where we used a Pioneer 3 robot to do the parallel parking maneuver.
  1. Automatic Robot Parallel Parking, CSCI5551 project together with R. Sivaligam and F. Lewis, supervised by Prof. S. Roumeliotis

The views and opinions expressed in this page are strictly those of the page author.
The contents of this page have not been reviewed or approved by the University of Minnesota.