Cluster Analysis Investigation of Video Streams and Archives

• Activity Number: 311
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Brett Amidan*+ and Harold Trease
• Companies: Pacific Northwest National Laboratory and Pacific Northwest National Laboratory
• Keywords: Video Stream ; Video Archive ; Images ; Cluster Analysis

Abstract

Video data represent high-volume, highly redundant, and low-information content media. Many video archives or streams like surveillance video, webcams, and UAV video do not have additional accompanying audio or meta-data which could also be used for analysis. In these cases with only video present, extracting information through the use of automated computer based tools is very difficult. Processes and algorithms that extract data from video and then cluster the data are useful in populating a database of objects of interest to be used for monitoring real-time video feeds or searching archived data for similar objects. This paper discusses the transformation of video into high-dimensional feature data. It also will compare various clustering algorithms and their effectiveness in clustering similar objects and in processing large amounts of high-dimensional data.