CN108010066B - Multi-hypothesis tracking method based on infrared target gray level cross-correlation and angle information - Google Patents

Abstract

The invention relates to a multi-hypothesis tracking method based on infrared target gray-scale cross-correlation and angle information, which reconstructs a track confidence evaluation model in the multi-hypothesis tracking method by utilizing the infrared target gray-scale cross-correlation characteristic information and the angle information, establishes a track confidence evaluation model based on the gray-scale cross-correlation and the angle information, completes track confidence calculation by utilizing the model, realizes the quick confirmation and quick deletion of multi-target tracks, reduces the number of subsequent track hypotheses, effectively improves the track establishment time and reduces the algorithm complexity. Therefore, the track confidence evaluation model based on the infrared target gray level cross correlation and the angle information can achieve a good effect, and is superior to a multi-hypothesis tracking method using the angle information track evaluation model.

Description

Multi-hypothesis tracking method based on infrared target gray level cross-correlation and angle information

Technical Field

The invention belongs to the technical field of data processing, relates to a method for multi-target tracking, and particularly relates to a multi-hypothesis tracking method based on infrared target gray level cross-correlation and angle information.

Background

Because the amount of information that the infrared alarm system can obtain is less, only target angle information generally exists, and there are a large amount of false alarms and clutter in the target information obtained, there is a contradiction between real-time and accuracy in multi-target tracking, need carry on its further study. At present, the generalized multi-target tracking algorithm mainly comprises a Global Nearest Neighbor (GNN) method, a joint probability data interconnection (JPDA) method and a multi-hypothesis tracking Method (MHT). The global nearest neighbor method and the joint probability data interconnection method are small in calculation amount and high in instantaneity, but targets are easy to lose in a high-density clutter environment; the multi-hypothesis tracking algorithm is a method for solving the problem of highest correlation accuracy of multi-target data in a high clutter environment, but the algorithm is large in calculation amount and poor in real-time performance.

Disclosure of Invention

Technical problem to be solved

In order to avoid the defects of the prior art, the invention provides a multi-hypothesis tracking method based on infrared target gray-scale cross-correlation and angle information, which reconstructs a track confidence evaluation model in multi-hypothesis tracking by using the infrared target gray-scale cross-correlation characteristics and the angle information, realizes the quick confirmation and quick deletion of multi-target tracks, effectively improves the track establishment time and reduces the algorithm complexity.

Technical scheme

A multi-hypothesis tracking method based on infrared target gray level cross correlation and angle information is characterized in that: firstly, initializing a flight path, and then, the steps are as follows:

step 1: performing data association, performing Kalman k times of filtering on the predicted flight path, and performing residual error of a filter

Residual covariance matrix S and residual norm d²The following were used:

S＝HP(k/k)H^T+R(k)

in the formula, H is a measurement matrix,

for the filter output, P (k/k) is an error covariance matrix, R (k) is a measurement noise covariance matrix, c is an associated threshold, and a self-adaptive threshold is adopted;

step 2, track evaluation: calculating a track confidence, which is generated by recursion accumulation, and the track confidence is as follows:

L(k)＝L(k-1)+ΔL_k

wherein: p_dProbability of detection for the target, beta_fFor false alarm spatial detection density, M is the measurement spatial dimension, d²Is the residual norm, S is the residual covariance matrix, Cov (k | k-1) is the cross-correlation coefficient of Track (k-1) associated target 5 x 5 neighborhood with Track (k) associated target 5 x 5 neighborhood;

the above-mentioned

Step 3, track deletion:

when L (k) < T_LDeleting the flight path;

when L (k) > T_UConfirming the flight path;

wherein, T_LPruning confidence thresholds, T, for a track_UConfirming a confidence threshold for the flight path;

when T is_L＜L(k)＜T_U：

1. And (3) track clustering: and (3) track clustering generation criterion: no interaction in different track classes;

2. and (3) assuming a track: track hypothesis generation criteria: traversing all possible tracks in the track cluster; optionally selecting one flight path as a hypothetical flight path; excluding all tracks that are not compatible with this track;

calculating the confidence coefficient of the assumed flight path; repeating the process from the flight path clustering to the flight path hypothesis;

step 4, track prediction: and performing Kalman filtering prediction on the confirmed track by adopting a uniform acceleration model CA to complete track filtering prediction, wherein the formula is as follows:

in the formula:

wherein X (k-1) is a target state vector, A (k-1) and E (k-1) respectively represent target azimuth angle and pitch angle data at the k-1 moment,

respectively represents the target position and the pitching motion speed at the k-1 moment,

respectively representing the target position and the pitching motion acceleration at the k-1 moment; f (k-1) is a state transition matrix, and G (k-1) is an input matrix; t is sampling time of a detector of the infrared alarm system; w (k-1) is state noise, is zero mean Gaussian white noise, and has a covariance matrix of Q (k-1);

and repeating the steps 1 to 4 for the next frame of target data.

Advantageous effects

The invention provides a multi-hypothesis tracking method based on infrared target gray-scale cross-correlation and angle information, which is characterized in that a track confidence evaluation model in the multi-hypothesis tracking method is reconstructed by utilizing the infrared target gray-scale cross-correlation characteristic information and the angle information, the track confidence evaluation model based on the gray-scale cross-correlation and the angle information is established, track confidence calculation is completed by utilizing the model, the rapid confirmation and the rapid deletion of multi-target tracks are realized, the number of subsequent track hypotheses is reduced, the track establishment time is effectively prolonged, and the algorithm complexity is reduced. Therefore, the track confidence evaluation model based on the infrared target gray level cross correlation and the angle information can achieve a good effect, and is superior to a multi-hypothesis tracking method using the angle information track evaluation model.

Drawings

FIG. 1 is a schematic flow chart of the method of the present invention

Detailed Description

The invention will now be further described with reference to the following examples and drawings:

the experimental data source of an embodiment includes a sequence of 200 frames of images (640 x 512) and corresponding detected target location data.

The method comprises the following specific implementation steps:

1) data association

And the data association adopts the predicted flight path and is filtered for k times. Computing residual of a filter according to kalman filtering principles

Residual covariance matrix S and residual norm d²：

S＝HP(k/k)H^T+R(k) (2)

In the formula, H is a measurement matrix,

for the filter output, P (k/k) is the error covariance matrix, R (k) is the measured noise covariance matrix, c is the correlation threshold,an adaptive threshold is used.

2) Track evaluation

First, Track Track (k-1) is taken in 5 x 5 neighborhood x of k-1 times associated target_i,j(k-1) (i is more than or equal to 1 and less than or equal to 5, j is more than or equal to 1 and less than or equal to 5), track picking (k) is associated with the target for k times in 5 x 5 neighborhoods x_i,j(k) (i is more than or equal to 1 and less than or equal to 5, and j is more than or equal to 1 and less than or equal to 5), the gray level cross correlation coefficient Cov (k | k-1) is as follows:

in the formula (I), the compound is shown in the specification,

calculating a track confidence, which is generated by recursion accumulation, and the track confidence is as follows:

L(k)＝L(k-1)+ΔL_k (5)

in the formula: p_dProbability of detection for the target, beta_fFor false alarm spatial detection density, M is the measurement spatial dimension, d²Is the residual norm, S is the residual covariance matrix, Cov (k | k-1) is the cross-correlation coefficient of Track (k-1) associated target 5 x 5 neighborhood with Track (k) associated target 5 x 5 neighborhood.

3) Track pruning

Calculating the track confidence L (k) according to the formula (6), and according to the L (k) < T_LDeleting the flight path; l (k) > T_UConfirming the flight path; t is_L＜L(k)＜T_UAnd waiting for more data to change the criterion for track deletion. Wherein, T_LPruning confidence thresholds, T, for a track_UA confidence threshold is identified for the track.

T_L＜L(k)＜T_U:

1. Track clustering

And (3) track clustering generation criterion: there is no interaction in the different track classes.

2. Track hypothesis

Track hypothesis generation criteria: traversing all possible tracks in the track cluster; optionally selecting one flight path as a hypothetical flight path; excluding all tracks that are not compatible with this track;

calculating the confidence coefficient of the assumed track by using the formula (5); and repeating the 1. track clustering-2. track hypothesis.

4) Flight path prediction

And performing Kalman filtering prediction on the confirmed track by adopting a uniform acceleration model (CA), and completing the track filtering prediction by using a formula (7), wherein the formula is as follows:

in the formula:

wherein X (k-1) is a target state vector, A (k-1) and E (k-1) respectively represent target azimuth angle and pitch angle data at the k-1 moment,

respectively represents the target position and the pitching motion speed at the k-1 moment,

respectively representing the target position and the pitching motion acceleration at the k-1 moment; f (k-1) is a state transition matrix, and G (k-1) is an input matrix; t is infrared alarm system detector samplingTime; w (k-1) is state noise, zero mean white Gaussian noise, and its covariance matrix is Q (k-1).

And (5) receiving new target data, and repeating the steps 1) -5) to finish multi-target tracking.

Claims (1)

1. A multi-hypothesis tracking method based on infrared target gray level cross correlation and angle information is characterized in that: firstly, initializing a flight path, and then, the steps are as follows:

step 1: performing data association, performing Kalman k times of filtering on the predicted flight path, and performing residual error of a filter

Residual covariance matrix S and residual norm d²The following were used:

S＝HP(k/k)H^T+R(k)

in the formula, H is a measurement matrix,

for the filter output, P (k/k) is the error covariance matrix, R (k) is the measurement noise covariance matrix;

step 2, track evaluation: calculating a track confidence, which is generated by recursion accumulation, and the track confidence is as follows:

L(k)＝L(k-1)+ΔL_k

wherein: p_dProbability of detection for the target, beta_fFor false alarm spatial detection density, M is the measurement spatial dimension, d²Is the residual norm, S is the residual covariance matrix, Cov (k | k-1) is the cross-correlation coefficient of Track (k-1) associated target 5 x 5 neighborhood with Track (k) associated target 5 x 5 neighborhood;

the above-mentioned

Step 3, track deletion:

when L (k) < T_LDeleting the flight path;

when L (k) > T_UConfirming the flight path;

wherein, T_LPruning confidence thresholds, T, for a track_UConfirming a confidence threshold for the flight path;

when T is_L＜L(k)＜T_U：

1) Track clustering: and (3) track clustering generation criterion: no interaction in different track classes;

2) track assumption: track hypothesis generation criteria: traversing all possible tracks in the track cluster; optionally selecting one flight path as a hypothetical flight path; excluding all tracks that are not compatible with this track;

calculating the confidence coefficient of the assumed flight path; repeating the process from the flight path clustering to the flight path hypothesis;

step 4, track prediction: and performing Kalman filtering prediction on the confirmed track by adopting a uniform acceleration model CA to complete track filtering prediction, wherein the formula is as follows:

in the formula:

wherein X (k-1) is a target state vector, A (k-1) and E (k-1) respectively represent target azimuth angle and pitch angle data at the k-1 moment,

respectively represents the target position and the pitching motion speed at the k-1 moment,

respectively representing the target position and the pitching motion acceleration at the k-1 moment; f (k-1) is a state transition matrix, and G (k-1) is an input matrix; t is sampling time of a detector of the infrared alarm system; w (k-1) is state noise, is zero mean Gaussian white noise, and has a covariance matrix of Q (k-1);

and repeating the steps 1 to 4 for the next frame of target data.

Images