0.2 Feature extraction from cs data

Can random noise yield specific information?

The novel challenge with using random basis projections for intelligent motion detection is that there is nospatial information about the image or movie in the compressed sensing data. Traditional pixel-based cameras provide a graph oflight intensity over position and, logically, most pixel-based detection approaches use the information about where the motionoccurs to help classify it. [1] The CS data provides us not withintensity at a point, but with the similarity (inner product) between the original pixel image and a selection of basis elementscomposed of random noise spread throughout the image plane. Intelligent detection for CS, therefore, must use approachesradically different from detection used on conventional video.

Simplicity for low power

A key feature of CS systems is the potential for extremely low power consumption. To keep the overall power ofthe system low, any computations we perform must be low power as well. For practical application, the algorithms chosen must be simple to compute.

Investigation goals

Working in a very new field, we began our project with the open-ended goal of researching what types ofmotion can be detected using compressed sensing and of implementing at least one such measurement. After much deliberation and trial,we chose to implement speed detection, specifically calculating the speed of a known object along the direction of motion. (Differentshapes complicate the problem, as we will discuss.) Looking ahead to more sophisticated detection, we also wanted an extensiblesystem to use the results of several relevant calculations to automatically characterize motion.

Though beyond the scope of our investigation, our motivation is a camera running our intelligent detection systemcapable of determining if a desired type of motion is observed. We pursue algorithms which could be implemented in real time at thedata collection point.

[1] For some interesting work in intelligentdetection with traditional cameras, see the work of James W. Davis and Aaron F. Bobick out of the MIT Media Laboratory in the late1990s.