Real Time Sensor Image Fusion

Period of Performance: 12/02/2002 - 12/02/2004

$725K

Phase 2 SBIR

Recipient Firm

Max-viz, Inc.
16165 SW 72
Portland, OR 97224
Principal Investigator

Abstract

The key to realization of an integrated Enhanced Vision System is the development of an economical and compact processor that can perform real-time adaptive, multi-imager fusion and ground-map correlation. We achieve statistically appropriate (Bayesian) operations in real time using a self-organizing, "association engine" neural network, implemented in reconfigurable-FPGA based hardware that optimally manages process/memory interface and capacity. In Phase I, it has been shown that image feature sets can be represented through emulation of the human visual (ventral) pathway; this involves imager preprocessing to obtain large, sparse, binary vectors including a temporal dimension (multiple-image-frames). The association engine acts as a weighted matrix that is "trained" to recognize the database (ground-map) features, and the approach has been shown to perform this as a Maximum Likelihood operation. In Phase II, the preprocessing will be finalized for all relevant sensors and the ultimate FPGA/ASIC requirements defined and implemented. The prototype processor will be refined using real sensor data, and ultimately demonstrated in real time flight tests. Pilot (human factors) interface, avionics (machine) interface, flight operational, and certification issues will be resolved in preparation for dual-use commercialization.