Title:
Video Analytics
Date:
Thursday, October 25, 2012 @ 1:00pm
Location:
E2-304 EITC Building, University of Manitoba Fort Garry Campus
Speaker:
Dr. Amir Meghdadi
Department of Computer Science
University of Manitoba
Abstract:
Many applications rely on extracting insight from large volumes of video.
One notable example includes video surveillance which generates data from over hundreds of video cameras at any given time (700 cameras exist campus wide). Consequently, identifying major events and browsing video content is a laborious and time consuming task and in dire need of novel automatic and interactive information seeking techniques. Video analytics, a nascent field, lies at the intersection of computer vision/algorithmic approaches to video content analysis and human-computer interaction. Video analytics is particularly useful in cases where search criteria are loosely defined or unstructured, such as identifying the root cause of vandalism, detecting suspicious behaviour or identifying outliers in routine video data.
In this talk I will describe my ongoing work in video analytics. My approach involves exploiting existing or developing novel computer vision techniques to extract events of interest, based on the operator’s input criteria. In my approach I focus on allowing users to extract insight from surveillance video records wherein moving objects (pedestrians, vehicles, animals) can be tracked over space and time. Extracting motion tracks using an efficient algorithm allows me to then summarize the video content using various visualization techniques, two of which I present in this talk. I implemented Action Shots that provide a stroboscopic representation of object movement in one still image. I also exploit the use of the Space-time Cube, a method for showing large amounts of tracks in a 3D image space with motion represented on the x-z plane and time on the y-axis. With Action Shots and Space-time Cube tracks, users can observe significantly more events and query for items of interest in specific regions of interest rapidly. I demonstrate my implementations of these visualizations in a tool sViSIT (selective Video Summarization and Interaction Tool) that also enables advanced querying capabilities. I provide a quick tour of features in sViSIT and discuss future research goals. I finally end the presentation with grand research themes for video analytics.
Cost:
This will be a free event.
Contact:
If you would like additional information or if you might be interested in presenting a seminar, please contact Stephane Durocher or the Department of Computer Science.
