Hughes Aircraft Company

Jacob M Sacks

Guy B Coleman

James W Henderson

LeRoy F Ichida

F Anthony Carter

Robert A Sturla

Operating at real-time data rates, the disclosed hardware network generates a signal which closely approximates the Mth-largest of a set of R input data signals. The operational basis of this network is a comparison (121-127) between the input data (101-107) and a monotonically-scanning reference function (110). At that point (140, 150, 160) when the lower (R-M+1) of the inputs have been equaled by an increasing reference, the reference has become the same as the Mth-largest and is used (119b, 165, 170, 175) as the filter output. Analogous operation is achieved with a decreasing reference. When the number R of inputs is odd and M is made equal to ((R+1)/2), the network becomes a real-time median filter.

Real-time ordinal-value filter utilizing reference-function comparison

A hardware logic network, operating at real-time data rates, generates a signal which matches the Mth-largest signal of an input data set. To accomplish this signal-matching, the network utilizes an iterative series of value adjustments which cause an intermediate working quantity to converge to a number whose rank in comparison to the input set is the same as that of the true Mth-largest. When test comparisons indicate that the intermediate value's relative rank is too low, the value-adjustment is upward. The adjustment is downward otherwise. Operating upon input data whose range of possible values is limited to a predetermined interval, the system takes as its initial working quantity the midpoint of this interval. The magnitude of the first adjustment is one-half of the remaining interval on either side of the initial approximation. Subsequent magnitude adjustments are in turn one-half of the preceeding half-remaining-interval magnitude. When the number R of inputs is odd and M is set equal to (R+1)/2, the network becomes a real-time median filter.

Real-time ordinal-value filter utilizing half-interval ranking

Operating at real-time data rates, the disclosed hardware apparatus determine which one of a set of R input data signals is the Mth-largest. Mutual comparisons between the data values themselves provides the basis for the mechanized determination schemes. After every data signal is pair-wise compared (112-167) with every other data signal, each subset of results (C12-C17, C12-C27, C13-C37, C14-C47, C15-C57, C16-C67), consistng of the outcome of the comparisons between a given data signal and all other data signals, is tested (210-260) to determine whether the results indicate that (M-1) of the other data signals are greater than the given signal. That data signal whose result set satisfies this (M-1) condition is the Mth-largest and is used (270, 280) as the network output (285). For data that is serially-presented (490,400) a network simplification makes set-testing possible after only one data value, that most-recently-presented, is compared (412-417) to all other (R-1) signals of the set. The simplification entails reducing the number of comparators utilized, while employing storage elements (512-567) to preserve for set-testing purposes the appropriate ones of the results of comparisons made prior to the presentation of the most-recent serial data element. When the number R of inputs is odd and M is made equal to (R+1)/2, the networks become real-time median filters.

Real-time ordinal-value filters utilizing complete intra-data comparisons

Operating at real-time data rates, hardware logic networks (FIGS. 3-6) receive onto a data-channel array (390 in FIG. 3) a set of unordered input data values and iteratively pair-wise transpose the set members until their positional order on the array coincides with the order the members would assume if their magnitudes were arranged according to sequential ordinal rank. Pair-wise comparisons between the data values themselves provide the basis for the mechanized pair-wise transposition schemes. Within each network, a key building block for performing both the pair comparisons and member transpositions is a Number Pair Orderer (NPO) (FIG. 2). Each NPO compares (210) two data elements and passes the smaller to one of its outputs (240), while passing the larger to the other of its outputs (260). Individual NPO's are arranged into two groups (311-315 and 321-325), each of which operates iteratively upon pairs of substantially all of the data values. The data-channel array, mechanized in a recirculatory configuration, carries the data back and forth between the two NPO groups. By making the pairs ordered by one NPO group mutually offset from the pairs ordered by the other group, a pair-to-pair transfer of unordered data elements to array positions beyond adjacent channels is accomplished. The recirculatory pair-wise ordering action is complete when the data has passed through the alternate NPO groups a sufficient number of times to enable a maximally-unordered set to be transposed into proper sequence. The networks become real-time median filters when configured to receive an odd number R of inputs and the ((R+1)/2)nd-largest member of the final ordered set is selected as the overall network output.

Real-time ordinal-value filters utilizing partial intra-data comparisons

A scene tracker for use with imaging or quasi-imaging sensors such as TV cameras, FLIRS, millimeter-wave scanners, or the like is disclosed. Specifically, the invention is concerned with a form of correlation tracking. At the start of a track mode, part or parts of a scene are stored in a memory. Every subsequent video scan is then compared with the scene or parts of the scene stored in memory to find the 'best match' between live data and stored data. This process generates signal output proportional to horizontal and/or vertical translations of the scene. Besides pure translation, the disclosed scene tracker is capable of tracking an original aim point in the presence of scene rotation, scene growth (zooming), scene reduction (dezooming), distortion, and/or partial scene obscuration. Special novel features of the invention include a multi-cell correlation principle; a novel method of computing and storing correlation functions for each cell; a method to derive weighted average displacements for the entities of all cell correlations; and a novel video signal processing technique.

Growing community of inventors

Thousand Oaks, CA, United States of America

Jacob M Sacks