Temporal compressive sensing for video


Video camera architects must design cameras capable of high-quality, dynamic event capture, while adhering to power and communications constraints. Though modern imagers are capable of both simultaneous spatial and temporal resolutions at micrometer and microsecond scales, the power required to sample at these rates is undesirable. The field of compressive sensing (CS) has recently suggested a solution to this design challenge. By exploiting physical-layer compression strategies, one may overlay the original scene with a coding sequence to sample at sub-Nyquist rates with virtually no additional power requirement. The underlying scene may be later estimated without significant loss of fidelity. In this chapter, we cover a variety of such strategies taken to improve an imager’s temporal resolution. Highlighting a new low-power acquisition paradigm, we show how a video sequence of high temporal resolution may be reconstructed from a single video frame taken with a low-framerate camera.