Low power image based triggering for extended operation surveillance

Abstract

It is desirable for many surveillance applications to have a portable high quality imaging system capable of continuous monitoring in remote locations, often for extended periods of time. Extended operation can be achieved with low power by taking advantage of the fact that no interesting action is occurring in the area of interest most of the time, allowing the camera to be turned off. This type of operation requires some type of trigger to detect when events occur and turn on the camera to collect imagery. A novel technique for this type of detection is the use of signal processing on low spatial and temporal resolution imagery with a low-power processor to detect action events. The low-resolution imager operation and low-power processor allow the system to consume minimal power, while still taking advantage of the information available from the imager. Triggering is done by performing background subtraction on the low resolution imagery to detect scene changes. Although there is extensive research on this subject, no known research has attempted to implement this type of algorithm in a low power system, which puts a significant constraint on the computation that can be performed. This paper describes research conducted at MIT Lincoln Laboratory to develop a power constrained background subtraction technique, and design a low power hardware system that utilizes this form of detection for image based triggering.