Image analysis – Applications – Target tracking or detecting
Reexamination Certificate
1999-08-04
2003-03-04
Johns, Andrew W. (Department: 2621)
Image analysis
Applications
Target tracking or detecting
Reexamination Certificate
active
06529614
ABSTRACT:
BACKGROUND OF THE INVENTION
The invention is a hybrid optoelectronic neural object recognition system (HONORS) comprising two major building blocks: (1) an advanced grayscale optical correlator (OC) and (2) a massively parallel neural three-dimensional neural-processor (N3DP).
To date, one of the most successful system architectures developed for optical processing is the optical correlator. A high-speed updatable optical correlator usually employs two spatial light modulators (SLMs), one for the input image and one for a correlator filter. Optical correlators have been developed for various pattern recognition applications. Previously, limitations to the modulation capabilities of the spatial light modulator used for the correlator filter had restricted the filter encoding to binary phase-only filters (BPOFs) and phase-only filters (POFs). Both BPOFs and POFs construct edge-enhanced filters for high-pass matched spatial filtering. Although both BPOFs and POFs provide high discrimination, they lack distortion tolerance among the same class of input images under surveillance. Moreover, the BPOF and POF algorithms strictly limit the development of optimum composite filter algorithms for distortion invariant correlation for factoring background noise out of target of interest identification.
Lockheed Martin (LM) has developed a binary optical correlator for darpa transfer of optical processing to systems (TOPS) program. Also, Litton Data Systems has developed a miniaturized, rugged optical correlator (MROC) system for air force. Below is a chart indicating the input resolution, filter type and package size for each of these systems.
LOCKHEED MARTIN (LM)
LITTON
INPUT
BINARY (256 × 256)
BINARY (256 × 256)
FILTER
BINARY (256 × 256, 3000 H
BINARY (256 256, 3000 Hz
PACKAGING
10″ × 10″ × 4″
5″ × 4″ 1.5″
RUGGED/TEM INVARIANT
RUGGED/MINIATURIZED
Both LM and Litton correlators are binary systems, and thus, they are not suited for operation in a noisy, cluttered environment. A binary modulation scheme does not permit using a smart distortion invariant filter algorithm such as that of the present invention described below. Nor does a binary modulation scheme provide the grayscale optical capability needed for distortion invariant, or noise filtering, automated target recognition or operation in cluttered & noisy environments.
SUMMARY OF THE INVENTION
It is an object of the present invention to solve the deficiencies of current systems noted above.
It is another object of the present invention to provide a hybrid optoelectronic neural object recognition system comprising an advanced grayscale optical correlator and a massively parallel neural 3-D processor.
It is another object of the present invention to provide an optical correlator which enables a 1000 frames/second speed as opposed to two frames/second which is provided by a digital implementation.
It is another object of the present invention to provide a compact, high-speed grayscale optical correlator for Ballistic Missile Defense Organization (BMDO) real-time automatic target recognition (ATR) applications.
It is another object of the present invention to provide a robust optical correlation filter algorithm for distortion invariant automatic target recognition.
It is another object of the present invention to provide an advanced miniature grayscale optical correlator.
It is another object of the present invention to provide a maximum average correlation height (MACH) filter for use in the gray scale optical correlator which generates an output correlation plan that has a high-peak response with respect to all input images from the same class.
It is another object of the present invention to provide a maximum average correlation height filter that maximizes the height of the mean correlation peak relative to the expected distortion of a target of interest (TOI) (e.g. scale, rotation, perspective angle, etc.).
It is another object of the present invention to provide superior performance of the maximum average correlation height filter attributed to its unique capability of reducing the filters' sensitivity to distortions and the removal of hard constraints on the peak of the signal, which permits the optimization of performance criterion for the optical correlator.
It is another object of the present invention to provide an optical correlator that is uniquely suitable for an optical implementation using the grayscale maximum average correlation height filter.
It is another object of the present invention to provide a new complex-valued spatial light modulation architecture for advanced automatic target recognition applications.
These and other objectives are met by a hybrid optoelectronic neural object recognition system (HONORS), a robust pattern recognition system, for high-speed detection, segmentation, classification and identification of objects from noisy/cluttered backgrounds.
The hybrid optoelectronic neural object recognition system consists of an optical correlator (OC) and a neural three-dimensional neural-processor (N3DP). The optical correlator consists mainly of a unique gray-scale spatial light modulator (SLM) as the high resolution correlation filter and is used for object detection and segmentation. Due to the inherent parallel processing capability of optical correlator, it performs wide area surveillance (up to 1000×1000 pixels per frame) of targets of interest (TOI) in less than 1 millisecond per frame.
The detected and segmented target of interest is then handed over to the three-dimensional neural-processor. The three-dimensional neural-processor consists of a 64-channel high-speed electronic convolver coupled to a multilayer electronic neural network. Each input (with a variable window size up to 64×64) is simultaneously mapped to 64 eigenvector-based object data bank images. The output from each input image is a 64-element feature vector. The electronic neural network subsequently classifies the input feature vector into multiple classes of object.
Both the correlation filter and the eigenimage data bank rely on training from example images of known classes. Training relies on rules developed using an optimization process. More specifically, a maximum average correlation height (MACH) algorithm is used for correlation filter training. Eigenimage computation is used to establish the object data bank.
Unique advantages of the hybrid optoelectronic neural object recognition system include: high-speed (<1 ms per frame), large input frame (up to 1000×1000), high discrimination accuracy (over 90%), and ease of training. The hybrid optoelectronic neural object recognition system could be critical to both Defense and NASA applications ranging from weapon and sensor platform guidance, wide area surveillance, spacecraft rendezvous and docking, navigation and guidance, landmark tracking for precision landing, and landing hazard avoidance.
With the recent advancement in spatial light modulator technology, bipolar-amplitude modulation is realizable using a Ferroelectric liquid crystal (FLC) spatial light modulator developed by Boulder Nonlinear System (BNS). The system of the present invention takes advantage of spatial light modulator technology, and adds a new and novel bipolar-amplitude filter (BAF) technique for optical correlator applications.
The grayscale input spatial light modulator is used to replace the binary spatial light modulator used in previous state-of-the-art optical correlator systems. This has eliminated the need of a binarization preprocessing step. A Ferroelectric Liquid Crystal (FLC) spatial light modulator (made by Boulder Nonlinear Systems) that is capable of encoding real-valued data is also used for the correlation filter implementation. This grayscale optical correlator has enabled the direct implementation of a gray-scale-based correlator composite filter algorithm for distortion invariant target/object detection and segmentation as required by the hybrid optoelectronic neural object recognition s
Chao Tien-Hsin
Daud Taher
Thakoor Anikumar
California Institute of Technology
Christie Parker & Hale LLP
Johns Andrew W.
Nakhjavan Shervin
LandOfFree
Advanced miniature processing handware for ATR applications does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Advanced miniature processing handware for ATR applications, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Advanced miniature processing handware for ATR applications will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3014505