Zack Merritt and Talon Ward, University of Central Florida


Object Tracking from Compressive Sensing Measurements without Inversion

Abstract: Compressive sensing is a recently popularized method of dimensionality reduction that uses the sparsity of a signal in some basis to recover the signal with only a small number of measurements. When sampling (an inner product of the vectorized original signal with a sensing matrix) is fast, compressive sensing can be used to accelerate a large, structured signal efficiently through a tight bottleneck. A common application is imaging. We explore methods of object tracking in a video signal from a compressive sensing camera without performing a CS inversion (i.e., without fully recovering the original signal). In particular, we focus on correlation tracking and demonstrate that tracking from the least energy solution introduces negligible error.

Mentor: (University of Central Florida)