The measurement efficiency of Compressive Sensing (CS) enables the computational construction
of images from far fewer measurements than what is usually considered necessary by the Nyquist–
Shannon sampling theorem. There is now a vast literature around CS mathematics and applications
since the development of its theoretical principles about a decade ago. Applications include quantum
information to optical microscopy to seismic and hyper-spectral imaging. In the application of
shortwave infrared imaging, InView has developed cameras based on the CS single-pixel camera
architecture. This architecture is comprised of an objective lens to image the scene onto a Texas
Instruments DLP® Micromirror Device (DMD), which by using its individually controllable
mirrors, modulates the image with a selected basis set. The intensity of the modulated image is then
recorded by a single detector.
While the design of a CS camera is straightforward conceptually, its commercial implementation
requires significant development effort in optics, electronics, hardware and software, particularly if
high efficiency and high-speed operation are required. In this paper, we describe the development of
a high-speed CS engine as implemented in a lab-ready workstation. In this engine, configurable
measurement patterns are loaded into the DMD at speeds up to 31.5 kHz. The engine supports
custom reconstruction algorithms that can be quickly implemented. Our work includes optical path
design, Field programmable Gate Arrays for DMD pattern generation, and circuit boards for front
end data acquisition, ADC and system control, all packaged in a compact workstation.
Obtaining high frame rates is a challenge with compressive sensing (CS) systems that gather measurements in a
sequential manner, such as the single-pixel CS camera. One strategy for increasing the frame rate is to divide the
FOV into smaller areas that are sampled and reconstructed in parallel. Following this strategy, InView has
developed a multi-aperture CS camera using an 8×4 array of photodiodes that essentially act as 32 individual
simultaneously operating single-pixel cameras. Images reconstructed from each of the photodiode measurements are
stitched together to form the full FOV.
To account for crosstalk between the sub-apertures, novel modulation patterns have been developed to allow
neighboring sub-apertures to share energy. Regions of overlap not only account for crosstalk energy that would
otherwise be reconstructed as noise, but they also allow for tolerance in the alignment of the DMD to the lenslet
array.
Currently, the multi-aperture camera is built into a computational imaging workstation configuration useful for
research and development purposes. In this configuration, modulation patterns are generated in a CPU and sent to
the DMD via PCI express, which allows the operator to develop and change the patterns used in the data acquisition
step. The sensor data is collected and then streamed to the workstation via an Ethernet or USB connection
for the reconstruction step. Depending on the amount of data taken and the amount of overlap between sub-apertures,
frame rates of 2–5 frames per second can be achieved. In a stand-alone camera platform, currently in
development, pattern generation and reconstruction will be implemented on-board.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.