Reliable Motion Detection, Location and Audit in Surveillance Video
Amirsaman Poursoltanmohammadi, Matthew Sorell · International Journal of Digital Crime and Forensics · 2009
The review of video captured by fixed surveillance cameras is a time consuming, tedious, expensive and potentially unreliable human process, but of very high evidentiary value. Two key challenges stand out in such a task: 1.) ensuring that all motion events are captured for analysis, and 2.) demonstrating that all motion events have been captured so that the evidence survives being challenged in court. In previous work (Zhao, Poursoltanmohammadi & Sorell, 2008), it was demonstrated that tracking the average brightness of video frames or frame segment provided a more robust metric of motion than other commonly hypothesized motion measures. This article extends that work in three ways: 1.) by setting automatic localized motion detection thresholds, 2.) by maintaining a frame-by-frame single parameter normalized motion metric, and 3.) by locating regions of motion events within the footage. A tracking filter approach is used for localized motion analysis, which adapts to localized background motion or noise within each image segment. When motion is detected, location and size estimates are reported to provide some objective description of the motion event.