Project Title:

Parallel Implementation of Image Correspondence Algorithms for Rotorcraft

03.04-1567
912258
Parallel Implementation of Image Correspondence Algorithms for Rotorcraft
Innovative Configuration, Inc.
9053 Soquel Drive, Suite 203
Aptos
CA
95003
Vason
Srini
408-688-1567
ARC
NAS2-13524
031

03.04-1567
912258

Abstract:

Parallel Implementation of Image Correspondence Algorithms for Rotorcraft
The operation of rotorcraft in high-threat environments requires assistance from
on-board computers, image sensors, and automation tools so that nap-of-the Earth
flight mode can be used to carry out a specified mission. The goal of this project
is to evaluate four parallel architectures for their ability to provide real-time
obstacle detection, range estimation, image correspondence, and other near-field
guidance calculations. The first architecture uses Datacube's MaxVideo 20 pipeline
image processing system, containing multiple functional units interconnected by a
32 x 32 crossbar. The second architecture uses Intel/CMU's iWarp chip interconnected
in a systolic manner. The third architecture involves general purpose microprocessors
such as Sparc interconnected using a shared bus to form a shared memory multiprocessor.
The fourth architecture employs special purpose, high-performance VLSI chips interconnected
in a systolic manner. The four architectures will be evaluated using the algorithms
and C programs available from NASA and other agencies during Phase I. These evaluations
will determine what portion of the parallelism present in the algorithms can be turned
into speedup by the architecture. The features of each of the four architectures
such as communication network, speed of ALU and multiplier, and pipelining that can
facilitate the exploitation of parallelism into speedup will be analyzed and documented.
Based on these evaluations, one of the architectures will be selected for implementation
in Phase II. The implementation will result in a board that can be plugged into the
SBUs slots of SparcStation 2 and experimented at NASA facilities.
The compact and low-cost board can be used in spacecraft and ships. The system can
also be used in full-motion video processing systems. The range estimation and image
correspondence algorithms will be applicable in controlling robots, monitoring flow
of materials in factories, and manufacturing.
rotorcraft flight, range estimation, image correspondence, kalman filters, digital
signal processor, video image processing, parallel system, systolic architectures