Moving vehicle classifier with iterative deconvolver estimator

• US 5,367,475 A
• Filed: 06/11/1993
• Issued: 11/22/1994
• Est. Priority Date: 09/23/1991
• Status: Expired due to Term

• Current Assignee: Conexant Systems Incorporated (Synaptics Incorporated)
• Original Assignee: Rockwell International Corporation
• Inventors: White, Stanley A.
• Primary Examiner(s): Mai, Tan V.
• Assistant Examiner(s): Ngo, Chuong D.
• Application Number:
  US08/076,077
• Time in Patent Office: 529 Days
• Field of Search: 364/724.01, 364/724.12, 364/724.19, 364/724.2, 364/728.01, 364/728.02, 375/11
• US Class Current: 708/300
• CPC Class Codes: G01S 7/412 based on a comparison between measured values and known or stored values
• CPC Class Codes: G01V 1/001 Acoustic presence detection
• CPC Class Codes: G08G 1/015 with provision for distinguishing between two or more types of vehicles, e.g. between motor-cars and cycles

Abstract

A MOVING VEHICLE CLASSIFIER WITH ITERATIVE DECONVOLVER ESTIMATOR deconvolves a signal in which a plurality of sensed signals, which originate from a common signal source (namely, a moving vehicle) and traverse differing signal paths, are used to formulate an initial unconstrained estimation of the signal source as a reference signal. Preselected constraints are placed on the reference signal to generate an estimate of the source signal, and of a source signal matrix, for the plurality of input sensors. A pseudoinverse of the source matrix is then used to form an estimate of the impulse response of the propagation paths for the signals. The estimates of signal path impulse response are used in combination with the input signals to again estimate the unconstrained reference signal. This is in turn used to estimate a new source signal matrix and path responses. These estimation steps are repeated in a series of iterative steps until a point of minimum variance from the received signals is reached. At this point, the source signal estimation is chosen as a deconvolved signal output. The moving vehicle which produced the source signal may then be classified by conventional means.

Claims

1. A method for classifying a moving vehicle which produces a source signal propagating along at least two signal paths, each path having an impulse response duration, the method comprising the steps of:

(a) sensing, at a minimum of one locations remote from the vehicle, a set of at least two received input signals propagated from the source signal along a minimum of two signal paths, and preprocessing such signals into input signal vectors xO, yO, . . . , zO, one vector from each sensor;

(b) collecting samples from each received input signal;

(c) for the set of received input signals;

(1) forming each of the received input signals into blocks of M+N-1 contiguous samples, where M is the duration of the source signal and N is the longest impulse response duration of any path, M and N both being measured in sampling intervals;

(2) measuring the energy of each block;

(3) selecting a block which has the most energy; and

(4) storing the input signal vectors, xO, yO, . . . , zO, received during said most energy block;

(d) combining a plurality of components of the input signal vectors, xO, yO, . . . , zO, by a root-sum-square technique, thereby producing a zeroth M+N-1 sample unconstrained reference signal vector, rO; and

setting an index i to have a value equal to one;

(e) selecting an (i-1)st unconstrained reference signal vector, r(i-1), for use in an ith iteration;

(f) forcing to zero all values of the (i-1)st unconstrained reference signal vector, r(i-1), except those which exceed a computed threshold, thereby producing an ith zeroed signal vector;

(g) discarding N-1 end samples of the ith zeroed signal vector such that the remaining M contiguous samples have more energy than any other grouping of M contiguous samples, thereby forming an ith constrained source signal vector, si;

(h) forming an tth constrained source signal matrix, [Si], from the ith constrained source signal vector, si, by setting each element of the ith constrained source signal matrix, [Si], at row p and column n to zero if n is greater than p or p is greater than M-1+n, otherwise setting that element equal to the (p-n)th element of the corresponding ith constrained source signal vector, si;

(i) premultiplying the ith constrained source signal matrix, [Si], by its transpose to form an ith product matrix, then postmultiplying the inverse of the ith product matrix by the transpose of the ith constrained source signal matrix, [Si], to form a pseudoinverse of the ith constrained source signal matrix, [Si];

(j) postmultiplying the pseudoinverse by each of the received input signal vectors, xO, yO . . . zO, to form an ith set of signal path impulse response estimation vectors, ci, di, . . . , ei;

(k) forming an tth set of signal path estimation matrices, [Ci], [Di], . . . , Ei, from the corresponding signal path impulse response estimation vectors, ci, di, . . . , ei, by setting each element of the ith signal path estimation matrices, [Ci], [Di9 . . . , Ei, at row p and column m to zero if m is greater than p or p is greater than N-1+m, otherwise setting that element equal to the (p-m)th entry of the corresponding signal path impulse response estimation vector, ci, di, . . . , ei;

(l) generating an tth set of input signal estimate vectors, xi, yi, . . . , zi, of the corresponding values of the input signal vectors, xO, yO, . . . , zO, by a step selected from the group consisting of the steps of;

(1) postmultiplying the ith constrained source signal matrix, [Si], by the corresponding signal path impulse response estimation vectors, ci, di, . . . , ei; and

(2) postmultiplying the corresponding ith set of signal path estimation matrices, [Ci], [Di], . . . , Ei, by the ith constrained source signal vector, si;

(m) obtaining an ith unconstrained reference signal vector, ri, by postmultiplying the inverse of an ith matrix [Mi] by a vector vi;

the ith matrix [[]Mi] consisting of the sum of all ith signal path estimation matrices, [Ci], [Di]. . . , Ei, premultiplied by their respective transposes; and

the vector vi consisting of the sum of all ith signal path estimation matrix transposes postmultiplied by their respective input signal vectors, xO, yO, . . . , zO;

(n) generating a first group of ith iteration estimation errors by subtracting from the input signal vectors, xO, yO . . . , zO, the corresponding input signal estimate vectors, xi, yi, . . . , zi;

(o) computing a first evaluation of estimator ith iteration performance by summing the squares of all scalar components of the first group of ith iteration estimation errors;

(p) generating a second group of ith iteration estimation errors by subtracting from the input signal vectors, xO, yO . . . , zO, the corresponding ith signal path estimation matrices, [Ci], [Di]. . . , Ei, each postmultiplied by the ith unconstrained reference signal vector, ri;

(q) computing a second evaluation of estimator ith iteration performance by summing the squares of all scalar components of said second group of ith iteration estimation errors;

(r) terminating the procedure if at least one of the first and second evaluations of the ith iteration performance indicate no further reduction in estimation errors, otherwise incrementing the value of i by one and repeating steps (e) through (r);

(s) selecting the ith constrained source signal vector, si, as the deconvolved source signal; and

(t) transferring the deconvolved source signal to an analysis apparatus.

2. The method of claim 1 wherein the step of forming the ith constrained source signal vector, si, further comprises the steps of:

(a) comparing the (i-1)st unconstrained reference signal vector, r(i-1), with a plurality of prestored sample source signal vectors; and

(b) selecting one of the prestored sample source signal vectors as a best fit match to the (i-1)st unconstrained reference signal vector, r(i-1), as the ith constrained source signal vector, si.

3. The method of claim 1 wherein the step of forming a constrained reference signal further comprises the steps of:

(a) performing peak picking on the (i-1)st unconstrained reference signal vector, r(i-1), so as to estimate locations of pulse peaks in the constrained source signal vector, si; and

(b) generating an output pulse sequence having pulses spaced apart relative to each other corresponding to locations for the pulse peaks.

4. The method of claim 1, wherein the procedure terminating step (r) is conditioned on both of the first and second evaluations of the ith iteration performance indicating no further reduction in estimation errors.

5. The method of claim 4 wherein the step of forming the ith constrained source signal vector, si, further comprises the steps of:

(a) comparing the (i-1)st unconstrained reference signal vector, r(i-1), with a plurality of prestored sample source signal vectors; and

(b) selecting one of the prestored sample source signal vectors as a best fit match to the (i-1)st unconstrained reference signal vector, r(i-1), as the ith constrained source signal vector, si.

6. The method of claim 4 wherein the step of forming a constrained reference signal further comprises the steps of:

(a) performing peak picking on the (i-1)st unconstrained reference signal vector. r(i-1). so as to estimate locations of pulse peaks in the constrained source signal vector, si; and

(b) generating an output pulse sequence having pulses spaced apart relative to each other corresponding to locations for the pulse peaks.