PhD in Electrical Engineering EECS Department University of Central Florida
I worked as a member of the Computer Vision Lab @ UCF. I’m working under Prof. Mubarak Shah.
Contact: ramin_not_ @ _not_cs.ucf.edu (remove _not_)
Latest News:
- Graduated on Dec 15th 2011.
- Defended my PhD dissertation on Nov 1st 2011.
- 2029 Downloads of the streakline code in 2 months.
- Internship in summer 2010, Transport and Mobility lab at EPFL, Lausanne, Switzerland, working under Dr. Michel Bierlaire
- Paper accepted for ECCV 2010, Crete, Greece, A streakline representation of flow in crowded scenes
- Attended the 3rd International summer school for computer vision (ICVSS 2010) in Sicily.
- Paper accepted for CVPR 2010, San Francisco, Human Identity Recognition
- Paper accepted for CVPR 09, Abnormal Crowd Behavior Detection using Social Force Model
-
ECCV 2010
Ramin Mehran, Brian Moore, Mubarak Shah,
A Streakline Representation of Flow in Crowded Scenes,
European Conference on Computer Vision (ECCV), Crete, Greece, 2010.
[PDF] [Project Page][code]
-
CVPR 2009
Ramin Mehran, Alexis Oyama, Mubarak Shah,
Abnormal Crowd Behavior Detection using Social Force Model,
IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), Miami, 2009.
[PDF] [Project Page]
-
CVPR 2010
Omar Oreifej, Ramin Mehran, and Mubarak Shah.
Human Identity Recognition in Aerial Images,
IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), San Francisco 2010.
[PDF][Project Page]
Research Projects
-
Shot Classification
Shot Classification based on Cinematographic Concepts, ALADDIN Project, 2011
We investigated a new method to recognize cinematographic concepts in videos based on camera motion. We used concepts from Lie Algebra to recognize 9 different shot types: Aerial, Brideye, Crane, Dolly, Establishing, Pan, Tilt, and Zoom. This approach has been applied to editorial videos as well as unconstrained videos from YouTube. We demonstrate that our approach outperforms the state of the art. I developed the method to use the Lie Algebraic representation of homography for shot classification.
-
Image Descriptor for Genre Recognition
Movie Genre Classification , ALADDIN Project, 2011
We investigate a novel image descriptor to integrate scene-level information of videos in a fast and effect manner. We classify videos based on bag of words approach using this descriptor. We demonstrate that our simple approach performs in par with CENTRIST feature on scene recognition in static images in 15 scene dataset and outperforms it for genre classification on a dataset of 264 movie trailers in four categories: Action, Comedy, Drama, and Horror.
-
Behavior Recognition in Moving Cameras
Behavior Recognition in Moving Cameras, Army Research Office, 2011
We use concepts from Lie Algebra to model the crowd motion patterns in videos from moving cameras and specially UAVs or helicopters. A novel framework for behavior recognition has been developed based on affine motion model of dense trajectories using Lie Algebriac representation.
-
Group Behavior Recognition
Group Behavior Recognition, Army Research Office, 2011
We investigate a method to model the group behaviors of humans. We look at the problem based on a part based method that models pair-wise and higher order interaction of motion patterns.
-
Data collection
Data Collection, NVESD, 2011
I led a group of five members for this project to collect a dataset of combatant and non-combatant personnel using two EO and one IR camera for pose and action recognition. We work on a novel method for pose recognition based on geometrical structure of SIFT matches.
-
Flow Representation in Crowded Scenes: Streaklines
Flow Representation in Crowded Scenes, Army Research Office, 2010
We borrowed the concept of streaklines from Fluid Mechanics and applied it to optical flows from videos of crowds. We demonstrate that the unique benefits of the new representation of the flow in motion segmentation, event detection, and behavior recognition.
-
Pedestrian Tracking using Discrete Choice Model
- Pedestrian Tracking using Discrete Choice Model , EPFL, 2010
We worked on a project to develop a method to mitigate the atmospheric turbulence in IR videos and detect and track moving objects. We developed a robust pipeline of non-rigid registration and object detection using a Gaussian model of the turbulence effect.
-
Turbulence Mitigation and Tracking
Turbulence Mitigation and Tracking in IR videos using KNIGHTS, NVESD, 2010
We worked on a project to develop a method to mitigate the atmospheric turbulence in IR videos and detect and track moving objects. We developed a robust pipeline of non-rigid registration and object detection using a Gaussian model of the turbulence effect.
-
Content based Search in Videos using Social Networks
Content based Search of Videos based on Social Interaction of Actors , IARPA, 2008-2009
We worked on content based search method for video by fusing three information modalities: visual concepts, social networks of the actors, and textual information. We developed a method to extract the emotional content of the text from movie scripts. In addition, we recognized the social network of the actors based on the co-appeared in visually similar scenes.
-
Event Detection using Social Force Model
Event Detection in Crowds based on Social Force Model, 2008-2009
We worked on developing a method for crowd event detection by using Social Force Model for the first time in the literature. We developed a holistic method for estimating the interaction force between individuals in the crowd using optical flow and dense particle advection technique. We use the estimated forces to learn a statistical model of the normal behaviors.