CN103942447B - Data fusion method and device for multi-source heterogeneous sensors - Google Patents

Abstract

The invention belongs to the technical field of data fusion, and provides a data fusion method and device for multi-source heterogeneous sensors. The method comprises the steps that measurement data of the various sensors are acquired, and time alignment is carried out on measurement data which are asynchronous in time and inconsistent in frequency through the spline fitting method; the measurement data obtained after alignment are grouped according to the classes and precision of the sensors; intra-group or inter-group combination is carried out on the grouped data, integrated measurement data of a target are provided, and preliminary estimation is carried out on the position of the target; linear transformation is carried out on a measurement equation with the preliminary estimation value of the position of the target as a reference point, and secondary estimation is carried out on the position of the target based on the weighted least square method. According to the data fusion method and device for the multi-source heterogeneous sensors, the three-layer fusion locating strategy is adopted, a monitoring system can acquire a high-precision target location value quickly, various kinds of target information collected by the acoustic/optical/electric sensors or other types of sensors at the front end of the system are fused more efficiently, and therefore the monitoring and commanding efficiency is improved.

Description

A kind of multi-source hybrid multisensor data fusion method and device

Technical field

The invention belongs to Data fusion technique field, particularly relate to a kind of multi-source hybrid multisensor data fusion method and device.

Background technology

Current, monitoring and commanding system is detecting means with visible ray/infrared composite photoelectric Detection Techniques, radio detection technology and the radar exploration technique etc. usually, take the tracking strategy of active/passive sensor phase mutual designation, confirmation mutually, form guarded region comprehensive situation.For realizing the seamless relay tracking to target in guarded region, need the metrical information for the multiple different systems such as photoelectricity, radar, radio in system, different sampling rate, propose a kind of data anastomosing algorithm of the Dissimilar sensors of multi-source fast and effectively system.

In data fusion, there is independent attainable in theory method the time alignment of measurement data and the location of target, conventional multi-source hybrid multisensor data fusion flow process as shown in Figure 1, front end sensors is generally incomplete measurement data, in multi-source Dissimilar sensors system, normally directly provide target state estimator value by cross bearing, will limited information be caused to be submerged in a large amount of false bearing point when system input increases, fusion results be had a greatly reduced quality.

Summary of the invention

In view of the above problems, the object of the present invention is to provide a kind of multi-source hybrid multisensor data fusion method and device, be intended to solve available data calculating fusion complexity high, locate inaccurate technical matters.

On the one hand, described multi-source hybrid multisensor data fusion method comprises the steps:

Obtain the measurement data of each sensor, and adopt spline-fitting method that measurement data inconsistent to time irreversibility, frequency is carried out time alignment, unified on a time point;

According to categories of sensors and precision, the measurement data after aligning is divided into groups;

One group of data that choice accuracy is higher from integrated data carry out organizing combination between interior combination and/or choice accuracy higher two groups of data groups, provide the perfect measurement data of target, carry out according to a preliminary estimate target location;

As a reference point with the first guess of target location, linear transformation is carried out to measurement equation, based on weighted least-squares method, quadratic estimate is carried out to target location, export quadratic estimate value.

On the other hand, described multi-source hybrid multisensor data fusion device comprises:

Time alignment module, for obtaining the measurement data of each sensor, and adopts spline-fitting method that measurement data inconsistent to time irreversibility, frequency is carried out time alignment, unified on a time point;

First Fusion Module, for dividing into groups to the measurement data after aligning according to categories of sensors and precision;

Second Fusion Module, carries out organizing combination between interior combination and/or choice accuracy higher two groups of data groups for one group of data that choice accuracy from integrated data is higher, provides the perfect measurement data of target, carry out according to a preliminary estimate target location;

3rd Fusion Module, as a reference point for the first guess with target location, linear transformation is carried out to measurement equation, based on weighted least-squares method, quadratic estimate is carried out to target location, export quadratic estimate value.

The invention has the beneficial effects as follows: the present invention is based on spline-fitting and imperfect measurement polishing method, propose multi-resources Heterogeneous fast hierarchical data anastomosing algorithm, reduce the complexity of target localization to a great extent; In addition, in Data-parallel language process, divide into groups and corresponding data assemblies according to different classes of carrying out, improve arithmetic accuracy, while location efficiency is merged in raising, optimize tracking effect further.By the present invention, supervisory system can obtain high-precision target localization value rapidly, and more efficiently all kinds of target informations that gather of the multiple sensors such as various sound/light/electricity of emerging system front end, improve the usefulness of control and command.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of existing multi-source hybrid multisensor data fusion method;

Fig. 2 is a kind of process flow diagram of the multi-source hybrid multisensor data fusion method that the embodiment of the present invention provides;

Fig. 3 is the angular altitude schematic diagram of 2D radargrammetry;

Fig. 4 is the distance schematic diagram of 2D passive sensor;

Fig. 5 is the schematic diagram of the cross bearing of 2D passive sensor;

Fig. 6 is that schematic diagram is combined in 2D passive sensor and laser ranging;

Fig. 7 is the another kind of process flow diagram of the multi-source hybrid multisensor data fusion method that the embodiment of the present invention provides;

Fig. 8 is the block diagram of the multi-source hybrid multisensor data fusion device that the embodiment of the present invention provides.

Embodiment

In order to make object of the present invention, technical scheme and advantage clearly understand, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, be not intended to limit the present invention.

In order to technical solutions according to the invention are described, be described below by specific embodiment.

embodiment one:

Fig. 2 shows the flow process of the multi-source hybrid multisensor data fusion method that the embodiment of the present invention provides, and illustrate only the part relevant to the embodiment of the present invention for convenience of explanation.

Step S21, obtain the measurement data of each sensor, and adopt spline-fitting method that measurement data inconsistent to time irreversibility, frequency is carried out time alignment, unified on a time point.

The measuring amount that sensor exports is a data sequence, because the frequency acquisition of different sensors is different with acquisition time interval, therefore the measurement data of each sensor output is asynchronous, and this step achieves carries out time alignment by the measurement data of different sensors, unified on a time point.Concrete comprises:

S211, for each sensor arrangement cubic spline functions, through spline interpolation matching, obtain a smooth curve.

Suppose certain sensor section [a sometime, b] in target carried out n+1 time measurement, whole time interval is divided into a=t0<t1< by sampling instant ... <tn=b, the observed reading that given moment point ti is corresponding is: f (ti)=yi (i=0,1, n), construct a cubic spline functions s (x), make it meet following condition:

①s(ti)＝yi，i＝0,1,…n；

2. s (t) is a cubic polynomial on each minizone [ti, ti+1], i=0,1 ... n;

3. s (t) has Second Order Continuous derivative on [a, b].

The construction process of cubic spline functions is as follows:

Note m _i=s ' (i=0,1,2 ..., n), at each minizone [t _i, t _i+1] (i=0,1 ... n), on, Hermite interpolation formula is utilized to write out the computing formula of cubic spline functions s (t):

$\begin{array}{c}s\left(t\right)=(1+2\frac{t-t_i}{t_{i+1}-t_i}){\left(\frac{t-t_{i+1}}{t_i-t_{i+1}}\right)}^2{y}_i+(1+2\frac{t-t_{i+1}}{t_i-t_{i+1}}){\left(\frac{t-t_i}{t_{i+1}-t_i}\right)}^2{y}_{i+1}\\ +(t-t_{i+1}){\left(\frac{t-t_{i+1}}{t_i-t_{i+1}}\right)}^2{m}_i(t-t_{i+1}){\left(\frac{t-t_i}{t_{i+1}-t_i}\right)}^2{m}_{i+1}\end{array}$

Utilize condition and subsidiary boundary condition, can system of equations be obtained:

$\left\{\begin{array}{c}{2m}_0+a_0{m}_1={\mathrm{\&beta;}}_0\\ (1-a_i)m_{i-1}+{2m}_i\\ (1-a_n)m_{n-1}+{2m}_n={\mathrm{\&beta;}}_n\end{array}+a_i{m}_{i+1}={\mathrm{\&beta;}}_i,(i=\mathrm{1,2},...,n-1)\right.$

By i=0,1,2 ..., n-1 substitutes into one by one, to solving equations, can draw recursion formula:

m _i＝a _im _i+1+b _i(i＝n,n-1,…,1,0)

Using formula asks b _i, a _i, make m _n+1=0, obtain m _n, m _n-1..., m ₀, by given parametric t _i, y _i, m _i(i=1,2 ..., n) substitute into s (x) and namely obtain required cubic spline functions.

Then a smooth curve is obtained by spline-fitting.Described spline-fitting is exactly at Spline Space S _kin (), to find out for f (t) about norm || || the best approach, namely find s ^*t (), makes through spline interpolation matching, a continuous print smooth curve can be obtained, can in the hope of sensor value at any time by this curve.

S212, to be as the criterion with the sampling instant of one of them sensor, from the smooth curve of other sensors, to take out the measured value in corresponding moment, realize time alignment.

Be as the criterion with a sampling instant, sampling instant be updated in the smooth curve of other sensors, the corresponding measured value that can ask, this results in the measurement data that each sensor is synchronous.

Step S22, according to categories of sensors and precision, the measurement data after aiming to be divided into groups.

Divide into groups to measurement data by categories of sensors and precision, the result after grouping can be divided into 2D radar measured data, 2D passive sensor data, laser ranging data etc.Described 2D passive sensor data can comprise infrared data, photooptical data, AIS data and ESM data etc.

Step S23, one group of data that choice accuracy is higher from integrated data carry out organizing combination between interior combination and/or choice accuracy higher two groups of data groups, provide the perfect measurement data of target, carry out according to a preliminary estimate target location.

Measurement data is divided into many groups, and extract one group or the higher measurement data of two groups of precision carries out cross bearing from each group, obtain perfect measurement data, preferably, the present embodiment comprises three kinds of array modes:

1) imperfect measurement data is supplemented by redundant information between 2D passive sensor data and 2D radar data.

2D passive sensor data and 2D radar data are incomplete data, and passive sensor data lack the distance of target, and the data of 2D radar lack the angular altitude of target.In order to supplement 2D radar data into partial data, need the angular altitude obtaining 2D radar.Angular altitude method for solving in 2D radar data is as follows: first by the measurement (sight line) of passive sensor and the planar S at 2D radar site determination target place _laS ₂, then the angular altitude of 2D radargrammetry is determined by the position angle of 2D radargrammetry and objective plane.As shown in Figure 3, if passive sensor position is respectively be measured as (β ₁, ε ₁) ^t, the position of 2D radar is be measured as (r ₂, β ₂) ^t.The then angular altitude ε of 2D radargrammetry ₂for

${\mathrm{\&epsiv;}}_2=\arctan \left\{\frac{\sin [{\mathrm{\&beta;}}_2-\arctan \left(\frac{y_{s_2}-y_{s_1}}{x_{s_2}-x_{s_1}}\right)]\tan {\mathrm{\&epsiv;}}_1+\frac{(z_{s_1}-z_{s_2})}{\sqrt{{(x_{s_2}-x_{s_1})}^2+{(y_{s_2}-y_{s_2})}^2}}\sin ({\mathrm{\&beta;}}_2-{\mathrm{\&beta;}}_1)}{\sin [{\mathrm{\&beta;}}_1-\arctan \left(\frac{y_{s_2}-y_{s_1}}{x_{s_2}-x_{s_1}}\right)]}\right\}---\left(1\right)$

The perfect measurement of the 2D radar after supplementing is (r ₂, β ₂, ε ₂) ^t.Polar coordinate measurement is converted into rectangular coordinate and is measured as

$\left\{\begin{array}{c}{x}_2=x_{s_2}+r_2\cos {\mathrm{\&epsiv;}}_2\cos {\mathrm{\&beta;}}_2\\ y_2=y_{s_2}+r_2\cos {\mathrm{\&epsiv;}}_2\sin {\mathrm{\&beta;}}_2\\ z_2=z_{s_2}+r_2\sin {\mathrm{\&epsiv;}}_2\end{array}\right.---\left(2\right)$

The measurement of 2D passive sensor lacks the distance of target.For solving the distance of target, a mistake (x can be made ₂, y ₂, z ₂) and the vertical line observing sight line vertical with 2D passive sensor, vertical line formation crossing with sight line intersection point X ₁, as shown in Figure 4, intersection point X ₁with sensor s ₁between distance can be used as the distance of target.Order

$\left\{\begin{array}{c}{p}_1=\cos {\mathrm{\&epsiv;}}_1\cos {\mathrm{\&beta;}}_1\\ q_1=\cos {\mathrm{\&epsiv;}}_1\sin {\mathrm{\&beta;}}_1\\ {\mathrm{\&gamma;}}_1=\sin {\mathrm{\&beta;}}_1\end{array}\right.---\left(3\right)$

The then distance r of target ₁for

$r_1=\frac{(x_2-x_{s_1})p_1+(y_2-y_{s_1})q_1+(z_2-z_{s_1}){\mathrm{\&gamma;}}_1}{p_1^2+q_1^2+{\mathrm{\&gamma;}}_1^2}=(x_2-x_{s_1})p_1(y_2-y_{s_1})q_1+(z_2-z_{s_1}){\mathrm{\&gamma;}}_1---\left(4\right)$

The perfect measurement of the 2D passive sensor after supplementing is (r ₁, β ₁, ε ₁) ^t, polar coordinate measurement is converted into rectangular coordinate and is measured as

$\left\{\begin{array}{c}{x}_1=x_{s_1}+r_1\cos {\mathrm{\&epsiv;}}_1\cos {\mathrm{\&beta;}}_1\\ y_1=y_{s_1}+r_1\cos {\mathrm{\&epsiv;}}_1\sin {\mathrm{\&beta;}}_1\\ z_1=z_{s_1}+r_1\sin {\mathrm{\&epsiv;}}_1\end{array}\right.---\left(5\right)$

Assuming that observation noise is separate Gaussian noise, then under rectangular coordinate system, the variance of target location is

$\left\{\begin{array}{c}{\mathrm{\&sigma;}}_{x_i}^2={\left(\frac{{\&PartialD;x}_i}{{\&PartialD;\mathrm{\&beta;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_1}^2+{\left(\frac{{\&PartialD;x}_i}{{\&PartialD;\mathrm{\&epsiv;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_1}^2+{\left(\frac{{\&PartialD;x}_i}{{\&PartialD;r}_2}\right)}^2{\mathrm{\&sigma;}}_{r_2}^2+{\left(\frac{{\&PartialD;x}_i}{{\&PartialD;\mathrm{\&beta;}}_2}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_2}^2\\ {\mathrm{\&sigma;}}_{y_i}^2={\left(\frac{{\&PartialD;y}_i}{{\&PartialD;\mathrm{\&beta;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_1}^2+{\left(\frac{{\&PartialD;y}_i}{{\&PartialD;\mathrm{\&epsiv;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_1}^2+{\left(\frac{{\&PartialD;y}_i}{{\&PartialD;r}_2}\right)}^2{\mathrm{\&sigma;}}_{r_2}^2+{\left(\frac{{\&PartialD;y}_i}{{\&PartialD;\mathrm{\&beta;}}_2}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_2}^2\\ {\mathrm{\&sigma;}}_{z_i}^2={\left(\frac{{\&PartialD;z}_i}{{\&PartialD;\mathrm{\&beta;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_1}^2+{\left(\frac{{\&PartialD;z}_i}{{\&PartialD;\mathrm{\&epsiv;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_1}^2+{\left(\frac{{\&PartialD;z}_i}{{\&PartialD;r}_2}\right)}^2{\mathrm{\&sigma;}}_{r_2}^2+{\left(\frac{{\&PartialD;z}_i}{{\&PartialD;\mathrm{\&beta;}}_2}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_2}^2\end{array}\right.---\left(6\right)$

In formula, with be respectively the position angle variance of sensor, angular altitude variance and range finding variance.

2) deficiency of data is supplemented by the redundant information between 2D passive sensor data.

The data of 2D passive sensor are sight line, and two passive sensors are measured and all lacked target range.For solving target range, what can utilize two sight lines vertical line public with it intersects to form two intersection point X ₁and X ₂.As shown in Figure 5.The distance of intersection point and respective sensor can be used as the distance of target.

If the coordinate of 2 passive sensors is respectively (β is observed to aerial same target ₁, ε ₁) ^t, (β ₂, ε ₂) ^t, order:

$\left\{\begin{array}{c}{p}_i=\cos {\mathrm{\&epsiv;}}_i\cos {\mathrm{\&beta;}}_i\\ q_i=\cos {\mathrm{\&epsiv;}}_i\sin {\mathrm{\&beta;}}_i\\ {\mathrm{\&gamma;}}_i={\sin \mathrm{\&epsiv;}}_i\end{array},(i=\mathrm{1,2})\right.---\left(7\right)$

Intersection point X _iwith sensor S _idistance r _ifor

$r_i=\frac{\left|\begin{array}{cc}{y}_{s_i}-y_{s_j}& z_{s_i}-z_{s_j}\\ q_j& l_j\end{array}\right|\&times;\left|\begin{array}{cc}qj& l_j\\ q_i& l_i\end{array}\right|+\left|\begin{array}{cc}{z}_{s_i}-z_{s_j}& x_{s_i}-x_{s_j}\\ l_j& p_j\end{array}\right|\&times;\left|\begin{array}{cc}{l}_j& p_j\\ q_i& l_i\end{array}\right|\left|\begin{array}{cc}{x}_{s_i}-x_{s_j}& y_{s_i}-y_{s_j}\\ p_j& q_j\end{array}\right|\&times;\left|\begin{array}{cc}{p}_j& q_j\\ p_i& q_i\end{array}\right|}{{\left|\begin{array}{cc}{p}_i& q_i\\ p_j& q_j\end{array}\right|}^2+{\left|\begin{array}{cc}{q}_i& l_i\\ q_j& l_j\end{array}\right|}^2+{\left|\begin{array}{cc}{l}_i& p_i\\ l_j& p_j\end{array}\right|}^2}---\left(8\right)$

Wherein, i, j=1,2, i ≠ j.By supplementing the partial data (r obtaining 2D sensor _i, β _i, ε _i) ^t, i=1,2.Polar measurement is converted into being measured as of rectangular coordinate:

$\left\{\begin{array}{c}{x}_i=x_{s_i}+r_i\cos {\mathrm{\&epsiv;}}_i\cos {\mathrm{\&beta;}}_i\\ y_i=y_{s_i}+r_i\cos {\mathrm{\&epsiv;}}_i\sin {\mathrm{\&beta;}}_i,i=\mathrm{1,2}\\ z_i=z_{s_i}+r_i\sin {\mathrm{\&epsiv;}}_i\end{array}\right.---\left(9\right)$

Under rectangular coordinate system, the variance of target location is:

$\left\{\begin{array}{c}{\mathrm{\&sigma;}}_{x_i}^2={\left(\frac{{\&PartialD;x}_i}{\&PartialD;{\mathrm{\&beta;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_1}^2+{\left(\frac{\&PartialD;x_i}{\&PartialD;{\mathrm{\&epsiv;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_1}^2+{\left(\frac{{\&PartialD;x}_i}{{\&PartialD;\mathrm{\&beta;}}_2}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_2}^2+{\left(\frac{{\&PartialD;x}_i}{{\&PartialD;\mathrm{\&epsiv;}}_2}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_2}^2\\ {\mathrm{\&sigma;}}_{y_i}^2={\left(\frac{\&PartialD;y_i}{\&PartialD;{\mathrm{\&beta;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_1}^2+{\left(\frac{\&PartialD;y_i}{\&PartialD;{\mathrm{\&epsiv;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_1}^2+{\left(\frac{\&PartialD;y_i}{\&PartialD;{\mathrm{\&beta;}}_2}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_2}^2+{\left(\frac{\&PartialD;y_i}{\&PartialD;{\mathrm{\&epsiv;}}_2}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_2}^2\\ {\mathrm{\&sigma;}}_{z_i}^2={\left(\frac{\&PartialD;z_i}{\&PartialD;{\mathrm{\&beta;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_1}^2+{\left(\frac{\&PartialD;z_i}{\&PartialD;{\mathrm{\&epsiv;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_1}^2+{\left(\frac{\&PartialD;z_i}{\&PartialD;{\mathrm{\&beta;}}_1}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_2}^2+{\left(\frac{\&PartialD;z_i}{\&PartialD;{\mathrm{\&epsiv;}}_2}\right)}^2{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_2}^2\end{array}\right.---\left(10\right)$

3) carry out combinations of pairs with 2D passive sensor data and laser ranging data and obtain perfect measurement data.

Suppose that stadimeter carries out target-seeking range finding along 2D passive sensor direction, if the coordinate of 2D passive sensor and stadimeter is respectively as shown in Figure 6, aerial same target be observed (β ₁, ε ₁) ^t, r ₂, then by supplementing the perfect measurement data obtained be:

$\left\{\begin{array}{c}{x}_i=x_{s_i}+(\sqrt{{(x_{s_1}-x_{s_2})}^2+{(y_{s_2}-y_{s_2})}^2+{(z_{s_1}-z_{s_2})}^2}+r_2)\cos {\mathrm{\&epsiv;}}_1\cos {\mathrm{\&beta;}}_1\\ y_i=y_{s_1}+(\sqrt{{(x_{s_1}-x_{s_2})}^2+{(y_{s_2}-y_{s_2})}^2+{(z_{s_1}-z_{s_2})}^2}+r_2)\cos {\mathrm{\&epsiv;}}_1\sin {\mathrm{\&beta;}}_1\\ z_i=z_{s_2}+(\sqrt{{(x_{s_1}-x_{s_2})}^2+{(y_{s_1}-y_{s_2})}^2+{(z_{s_1}-z_{s_2})}^2}+r_2)\sin {\mathrm{\&epsiv;}}_1\end{array}\right.$

By complete measurement data can be obtained to supplementing of imperfect measurement, thus obtain complete target location and the variance of target location.If the target bit obtained is equipped with M, be expressed as (xi, yi, zi), i=1,2 ..., the variance of M, M target location is adopt precision weighted method, the first guess of target location is:

${\hat{x}}_{*}=\frac{1}{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{1}{{\mathrm{\&sigma;}}_{x_i}^2}}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{x_i}{{\mathrm{\&sigma;}}_{x_i}^2},{\hat{y}}_{*}=\frac{1}{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{1}{{\mathrm{\&sigma;}}_{y_i}^2}}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{y_i}{{\mathrm{\&sigma;}}_{y_i}^2},{\hat{z}}_{*}=\frac{1}{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{1}{{\mathrm{\&sigma;}}_{z_i}^2}}\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}\frac{z_i}{{\mathrm{\&sigma;}}_{z_i}^2}---\left(11\right)$

Step S24, as a reference point with the first guess of target location, linear transformation is carried out to measurement equation, based on weighted least-squares method, quadratic estimate is carried out to target location, export quadratic estimate value.

With reference to Fig. 7, this step specifically comprises:

S241, by as a reference point for target location first guess, obtain sensor and measure equation accordingly;

S242, equation will be measured in the linearization of reference point place;

The weighted least-squares method of S243, utilization expansion carries out the quadratic estimate of target location.

In order to improve positioning precision, X as a reference point is estimated in the target location that formula (11) is tried to achieve _*, equation will be measured in the linearization of reference point place, and utilize the weighted least-squares method of expansion to carry out the quadratic estimate of target location.If X=(x, y, z) ^trepresent the position vector of target, represent sensor s _iposition vector.

1) if sensor s _ifor 2D passive sensor, its measurement equation is

$z_i=\left[\begin{array}{c}{\mathrm{\&beta;}}_i\\ {\mathrm{\&epsiv;}}_i\end{array}\right]=\left[\begin{array}{c}\arccos \left(\frac{x-x_{s_i}}{\sqrt{{(x-x_{s_i})}^2+{(y-y_{s_i})}^2}}\right)\\ \arctan \left(\frac{z-z_{s_i}}{\sqrt{{(x-x_{s_i})}^2+{(y-y_{s_i})}^2}}\right)\end{array}\right]+\left[\begin{array}{c}{v}_{{\mathrm{\&beta;}}_i}\\ v_{{\mathrm{\&epsiv;}}_i}\end{array}\right]=\left[\begin{array}{c}{f}_{{\mathrm{\&beta;}}_i}\left(X\right)\\ f_{{\mathrm{\&epsiv;}}_i}\left(X\right)\end{array}\right]+V_i---\left(12\right)$

Wherein V _ifor sensor s _imeasuring error matrix, and E (V _i)=0,

2) if sensor s _ifor 2D radar, its measurement equation is

$z_i=\left[\begin{array}{c}{\mathrm{\&beta;}}_i\\ r_i\end{array}\right]=\left[\begin{array}{c}\arccos \left(\frac{x-x_{s_i}}{\sqrt{{(x-x_{s_i})}^2+{(y-y_{s_i})}^2}}\right)\\ \sqrt{{(x-x_{s_i})}^2+{(y-y_{s_i})}^2+{(z-z_{s_i})}^2}\end{array}\right]+\left[\begin{array}{c}{v}_{{\mathrm{\&beta;}}_i}\\ v_{r_i}\end{array}\right]=\left[\begin{array}{c}{f}_{{\mathrm{\&beta;}}_i}\left(X\right)\\ f_{r_i}\left(X\right)\end{array}\right]+V_i---\left(13\right)$

Wherein V _ifor sensor s _imeasuring error matrix, and E (V _i)=0, the measuring set that N number of sensor forms is Z={z _i, i=1,2 ..., N}.

Utilize the location estimation that formula (11) is tried to achieve will x as a reference point _*, to formula (12), (13) linearization:

${\&PartialD;z}_i=z_i-z_{i*}=\frac{{\&PartialD;z}_i}{\&PartialD;X}{|}_{X=X_{*}}\&times;(X-X_{*})+V_i\frac{{\&PartialD;z}_i}{\&PartialD;X}{|}_{X=X_{*}}\&times;\mathrm{\&delta;X}+V_i=K_i{|}_{X=X_i}\&times;\mathrm{\&delta;X}+V_i---\left(14\right)$

To 2D sensor in formula $z_{i*}={\left[\begin{array}{c}{f}_{{\mathrm{\&beta;}}_i}\left(X\right)\\ f_{{\mathrm{\&epsiv;}}_i}\left(X\right)\end{array}\right]}_{X=X_{*}},$ To 2D radar $z_{i*}={\left[\begin{array}{c}{f}_{{\mathrm{\&beta;}}_i}\left(X\right)\\ f_{{\mathrm{\&epsiv;}}_i}\left(X\right)\end{array}\right]}_{X=X_{*}}.$ The measurement of N number of sensor is combined,

$\mathrm{\&delta;Z}=\left[\begin{array}{c}{z}_1-z_{1*}\\ z_2-z_{2*}\\ .\\ .\\ .\\ z_N-z_{N*}\end{array}\right]=\left[\begin{array}{c}{K}_1\\ K_2\\ .\\ .\\ .\\ K_N\end{array}\right]{|}_{X=X_{*}}\&times;\mathrm{\&delta;X}+\left[\begin{array}{c}{V}_1\\ V_2\\ .\\ .\\ .\\ V_N\end{array}\right]=K{|}_{X=X_{*}}\&times;\mathrm{\&delta;X}+V---\left(15\right)$

Make R=E (V _iv _i ^t), then can obtain the weighted least square of δ X

$\mathrm{\&delta;}\hat{X}={\left[{\left(K^T{R}^{-1}K\right)}^{-1}{K}^T{R}^{-1}\mathrm{\&delta;Z}\right]}_{X=X_{*}}---\left(16\right)$

Variance of estimaion error is

$P\left(\mathrm{\&delta;}\hat{X}\right)={\left[{\left(K^T{R}^{-1}K\right)}^{-1}\right]}_{X=X_{*}}---\left(17\right)$

The variance that can be obtained final estimated value and estimated value error by formula (16), (17) is

$\left\{\begin{array}{c}\hat{X}=X_{*}+\mathrm{\&delta;}\hat{X}\\ P\left(\hat{X}\right)=P\left(\mathrm{\&delta;}\hat{X}\right)\end{array}\right.---\left(18\right)$

S244, output quadratic estimate value are as target localization value.

Be further used as a kind of preferred implementation, described step S24 also comprises:

S240, by after as a reference point for target location first guess, the position of adjustment reference point.

The target location that formula (11) obtains is estimated to estimate under rectangular coordinate.And measurement is carried out under polar coordinate system.Under rectangular coordinate system, seem target location accurately estimate just may be forbidden as seen from polar coordinates.When target location is estimated inaccurate, can not directly use the location estimation of formula (11) as a reference point, the position of the whole reference point that now needs to wither.For passive sensor, if it measures (β _i, ε _i) ^tmeet formula (19), think the target location estimated inaccurate.

${\mathrm{\&lambda;}}_i=\frac{{[{\mathrm{\&beta;}}_i-f_{{\mathrm{\&beta;}}_i}\left({\hat{X}}_{*}\right)]}^2}{{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_i}^2}+\frac{{[{\mathrm{\&epsiv;}}_i-f_{{\mathrm{\&epsiv;}}_i}\left({\hat{X}}_{*}\right)]}^2}{{\mathrm{\&sigma;}}_{{\mathrm{\&epsiv;}}_i}^2}>\mathrm{\&lambda;}---\left(19\right)$

Now with reference to knock type (20) adjustment

$\left\{\begin{array}{c}{x}_{*}=x_{s_i}+\sqrt{{(x_{*}-x_{s_i})}^2+{(y_{*}-y_{s_i})}^2+{(z_{*}-z_{s_i})}^2}\cos {\mathrm{\&epsiv;}}_i\cos {\mathrm{\&beta;}}_i\\ y_{*}=y_{s_i}+\sqrt{{(x_{*}-x_{s_i})}^2+{(y_{*}-y_{s_i})}^2+{(z_{*}-z_{s_i})}^2}\cos {\mathrm{\&epsiv;}}_i\cos {\mathrm{\&beta;}}_i\\ z_{*}=z_{s_i}+\sqrt{{(x_{*}-x_{s_i})}^2+{(y_{*}-y_{s_i})}^2+{(z_{*}-z_{s_i})}^2}\sin {\mathrm{\&epsiv;}}_i\end{array}\right.---\left(20\right)$

For passive sensor, if it measures (β _i, ε _i) ^tmeet formula (19), think the target location estimated inaccurate.

${\mathrm{\&lambda;}}_i=\frac{{[{\mathrm{\&beta;}}_i-f_{{\mathrm{\&beta;}}_i}\left({\hat{X}}_{*}\right)]}^2}{{\mathrm{\&sigma;}}_{{\mathrm{\&beta;}}_i}^2}+\frac{{[r_i-f_{r_i}\left({\hat{X}}_{*}\right)]}^2}{{\mathrm{\&sigma;}}_{r_i}^2}>\mathrm{\&lambda;}---\left(21\right)$

Now with reference to knock type (22) adjustment

$\left\{\begin{array}{c}{x}_{*}=x_{s_i}+\sqrt{r_i^2-{(z_{*}-z_{s_i})}^2}\cos {\mathrm{\&beta;}}_i\\ y_{*}=y_{s_i}+\sqrt{r_i^2-{(z_{*}-z_{s_i})}^2}\sin {\mathrm{\&beta;}}_i\\ z_{*}={\hat{z}}_{*}\end{array}\right.---\left(22\right)$

In formula (19), (20), λ is threshold value.Due to λ _iobeying degree of freedom is the χ of 2 ²distribution, provides level of signifiance α=0.001, looks into χ ²distribution table obtains threshold value λ=13.816.

In this preferred implementation, introduce the reference point method of adjustment of band feedback mechanism, provide the adjustment process of 2 D radars and 2D passive sensor location estimation reference point respectively, be intended to improve estimated accuracy further.

embodiment two:

Fig. 8 shows the structure of the multi-source hybrid multisensor data fusion device that the embodiment of the present invention provides, and illustrate only the part relevant to the embodiment of the present invention for convenience of explanation.

The multi-source hybrid multisensor data fusion device that the present embodiment provides comprises:

Time alignment module 81, for obtaining the measurement data of each sensor, and adopts spline-fitting method that measurement data inconsistent to time irreversibility, frequency is carried out time alignment, unified on a time point;

First Fusion Module 82, for dividing into groups to the measurement data after aligning according to categories of sensors and precision;

Second Fusion Module 83, carries out organizing combination between interior combination and/or choice accuracy higher two groups of data groups for one group of data that choice accuracy from integrated data is higher, provides the perfect measurement data of target, carry out according to a preliminary estimate target location;

3rd Fusion Module 84, as a reference point for the first guess with target location, linear transformation is carried out to measurement equation, based on weighted least-squares method, quadratic estimate is carried out to target location, export quadratic estimate value.

Above-mentioned module 81-84 correspondence achieves the step S21-S24 in embodiment one.

Wherein, preferably, described time alignment module 81 comprises:

Spline-fitting unit 811, for for each sensor arrangement cubic spline functions, through spline interpolation matching, obtains a smooth curve;

Time synchronized unit 812, is as the criterion for the sampling instant with one of them sensor, takes out the measured value in corresponding moment, realize time alignment from the smooth curve of other sensors described.

Described 3rd Fusion Module comprises:

Reference point setting unit, for by as a reference point for target location first guess, obtain sensor and measures equation accordingly;

Linearizer, for measuring equation in the linearization of reference point place, will utilize the weighted least-squares method of expansion to carry out the quadratic estimate of target location;

Position output unit, for exporting quadratic estimate value as target localization value.

Preferably, described 3rd Fusion Module also comprises:

Reference point adjustment unit, for by after as a reference point for target location first guess, the position of adjustment reference point.

To sum up, this invention takes three layers of positioning strategy merged: first category divides into groups to sensor measurement data; Then, from grouped measures certificate, choose one or two measurement that precision is higher, carry out combining in group or between group, provide the perfect measurement value of target, target location according to a preliminary estimate; Last with the first guess of target location for reference point, carry out linear transformation to measurement equation, the weighted least-squares method based on expansion carries out the quadratic estimate of target location.Meanwhile, according to the tracking singularity not exclusively measuring Dissimilar sensors, take the Time Registration Method based on spline-fitting, the problem that solution data fusion time irreversibility, data transfer rate are inconsistent.

One of ordinary skill in the art will appreciate that, the all or part of step realized in above-described embodiment method is that the hardware that can carry out instruction relevant by program has come, described program can be stored in a computer read/write memory medium, described storage medium, as ROM/RAM, disk, CD etc.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any amendments done within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.

Claims (4)

1. a multi-source hybrid multisensor data fusion method, is characterized in that, described data fusion method comprises:

Obtain the measurement data of each sensor, and adopt spline-fitting method that measurement data inconsistent to time irreversibility, frequency is carried out time alignment, unified on a time point;

According to categories of sensors and precision, the measurement data after aligning is divided into groups;

One group of data that choice accuracy is higher from integrated data carry out organizing combination between interior combination and/or choice accuracy higher two groups of data groups, provide the perfect measurement data of target, carry out according to a preliminary estimate target location;

As a reference point with the first guess of target location, linear transformation is carried out to measurement equation, based on weighted least-squares method, quadratic estimate is carried out to target location, export quadratic estimate value;

The measurement data of wherein said each sensor of acquisition, and adopt spline-fitting method that measurement data inconsistent to time irreversibility, frequency is carried out time alignment, unified to step on a time point, specifically comprise:

For each sensor arrangement cubic spline functions, through spline interpolation matching, obtain a smooth curve;

Be as the criterion with the sampling instant of one of them sensor, from the smooth curve of other sensors, take out the measured value in corresponding moment, realize time alignment;

Wherein, described one group of data that choice accuracy is higher from integrated data carry out organizing combination between interior combination and/or choice accuracy higher two groups of data groups, provide the perfect measurement data of target, carry out step according to a preliminary estimate, specifically comprise target location:

Imperfect measurement data is supplemented by redundant information between 2 D radar datas and 2 D passive sensor data;

Imperfect measurement data is supplemented by the redundant information in 2 D passive sensor data groups;

Carry out combinations of pairs with 2D passive sensor data and laser ranging data and obtain perfect measurement data;

After obtaining perfect measurement data, target location is carried out according to a preliminary estimate;

Wherein, the described first guess with target location is as a reference point, carries out linear transformation, carry out quadratic estimate based on weighted least-squares method to target location to measurement equation, exports quadratic estimate value step, specifically comprises:

By as a reference point for target location first guess, obtain sensor and measure equation accordingly;

Equation will be measured in the linearization of reference point place, and utilize the weighted least-squares method of expansion to carry out the quadratic estimate of target location;

Export quadratic estimate value as target localization value.

2. multi-source hybrid multisensor data fusion method as claimed in claim 1, is characterized in that, measurement data comprises 2D radar data, 2D passive sensor data and laser ranging data after grouping.

3. multi-source hybrid multisensor data fusion method as claimed in claim 2, is characterized in that, after as a reference point for target location first guess, and the position of adjustment reference point.

4. a multi-source hybrid multisensor data fusion device, is characterized in that, described device comprises:

Time alignment module, for obtaining the measurement data of each sensor, and adopts spline-fitting method that measurement data inconsistent to time irreversibility, frequency is carried out time alignment, unified on a time point;

First Fusion Module, for dividing into groups to the measurement data after aligning according to categories of sensors and precision;

Second Fusion Module, carries out organizing combination between interior combination and/or choice accuracy higher two groups of data groups for one group of data that choice accuracy from integrated data is higher, provides the perfect measurement data of target, carry out according to a preliminary estimate target location;

3rd Fusion Module, as a reference point for the first guess with target location, linear transformation is carried out to measurement equation, based on weighted least-squares method, quadratic estimate is carried out to target location, export quadratic estimate value;

Wherein said time alignment module comprises:

Spline-fitting unit, for for each sensor arrangement cubic spline functions, through spline interpolation matching, obtains a smooth curve;

Time synchronized unit, is as the criterion for the sampling instant with one of them sensor, takes out the measured value in corresponding moment, realize time alignment from the smooth curve of other sensors described;

Wherein said 3rd Fusion Module comprises:

Reference point setting unit, for by as a reference point for target location first guess, obtain sensor and measures equation accordingly;

Linearizer, for measuring equation in the linearization of reference point place, will utilize the weighted least-squares method of expansion to carry out the quadratic estimate of target location;

Position output unit, for exporting quadratic estimate value as target localization value;

Wherein said 3rd Fusion Module comprises:

Reference point adjustment unit, for by after as a reference point for target location first guess, the position of adjustment reference point.