Abstract
Improving aviation safety is a primary mission of the National Aeronautics
and Space Administration (NASA).  One technology currently being developed
at NASA Langley to support this mission is the Enhanced Vision System
(EVS).  This system utilizes multi-sensor/multi-spectral image fusion to
provide enhanced images of the flight environment.  The enhanced images
are used to assist pilots in poor visibility conditions such as fog,
rain, snow or haze.  Extensive digital image processing is used to
generate the enhanced images.

Digital image processing is inherently computationally intensive.
Large amounts of data must be stored, processed and shuffled between
processor and memory.  Even simple point operations on a standard 640 x
480 8-bit color image required the manipulation of nearly a Megabyte (MB)
of data.  Other algorithm implementation issues such as non-sequential
data access or multiple line processing exaggerate the data processing
problems.  When video is considered, the data processing requirements
become enormous.  A single second of video data in the format mentioned
above recorded at 30 frames per second requires 30 MB of storage and
each frame must be processed within 33.3 milliseconds (ms) to sustain
the frame rate.

The digital image processing algorithm chosen to perform image enhancement
for the EVS is the Retinex algorithm.  The Retinex algorithm is a
general-purpose image enhancement algorithm for producing good visual
representations of scenes. It performs a non-linear spatial/spectral
transform that synthesizes strong local contrast enhancement and color
constancy.  It was developed and patented at NASA Langley.  The algorithm
was initially targeted to process multi-spectral satellite data but has
found applicability in very diverse areas such as medical radiography,
forensic investigations, consumer photography, and aviation safety.
The dynamic range compression, color constancy and sharpening provided
by the Retinex make it an ideal algorithm to use for enhancement of poor
weather images.

A real-time, video frame rate, implementation of the Retinex is required
to meet the needs of the EVS and other potential users. Retinex processing
contains a relatively large number of complex computations that, for
video exceed the performance capabilities of most general-purpose
computers.  In addition, several applications require an embedded,
low cost processor for computation.    To meet the goal of embedded,
real-time Retinex processing, we are currently studying and evaluating
the use of digital signal processors (DSPs) and reconfigurable logic
devices such as field programmable gate arrays (FPGAs) for implementation.
Preliminary results obtained using a DSP system indicate that real-time
operation is achievable.  This discussion will describe the EVS, Retinex
algorithm and DSP system currently under study.  Algorithm optimizations
and architecture trades used to achieve current performance levels will
also be presented.

Biography

Glenn Hines is a senior electronics engineer at NASA Langley Research
Center. He holds B.S and M.S. degrees in Electrical Engineering from
Old Dominion University and a M.S. degree in Computer Science from
the College of William and Mary. He is currently a Ph.D. candidate in
Computer Science at The College of William and Mary. He is responsible
for the design, development, and test of electronic systems for space
and aeronautic flight research instruments. He has developed application
specific integrated circuits (ASICs), multi-chip modules (MCMs), digital
signal processors (DSPs) and field programmable gate arrays (FPGAs) for
over 12 years.  He is currently performing research in the development and
optimization of real-time image enhancement algorithms and architectures.