CN103778629A - Background model real-time updating method for non-coherent radar image - Google Patents

Abstract

The invention discloses a method for real-time updating of a background model of a non-coherent radar image, which is used for low-altitude airspace security monitoring. The method of the present invention establishes a background model sample set for each pixel in the radar image, and compares the value of each pixel in the current frame image with the background model sample set of pixels in the current background model during the background model update process. The pixel value belongs to a sample of the background model, and a sample in the current background model is randomly replaced with this pixel value. At the same time, the background model sample set of a certain neighborhood pixel is randomly updated with the pixel value. The update cycle of the background model is also set randomly. The method of the invention prolongs the time that each sample stays in the background model set while keeping the number of background model samples constant, and broadens the time coverage of the background model samples.

Description

A method for real-time updating of background model in non-coherent radar images

technical field

The invention relates to a method for real-time updating of a background model of a non-coherent radar image, belongs to the technical field of low-altitude airspace security monitoring, and relates to radar image processing and target detection.

Background technique

Primary non-coherent radar has the characteristics of low cost, convenient installation, and strong independent workability, and is an important means of airspace security surveillance. Non-coherent radar itself does not have the function of moving target detection. A mature radar surveillance system usually uses an image acquisition card to transmit the radar plane position indication image to the computer, and then it is processed by the back-end image-based target detection algorithm to extract dispatch target information. Background subtraction technology is the most commonly used moving target detection technology. However, since the area monitored by the system is low-altitude airspace, the background environment is complex, the noise interference is strong, and the echo of background objects has large fluctuations and has certain random characteristics. Therefore, establishing an accurate background model and using an appropriate method to update it in real time becomes the key to improving the ability of object detection.

The traditional background model update method usually uses the newly extracted background pixel samples in the current frame image to replace the samples with the longest retention time in the original background model, causing each sample to remain in the background model for exactly the same time, which limits the background model. The temporal coverage of the sample.

Contents of the invention

The purpose of the present invention is to provide a real-time update method of a background model of a non-coherent radar image, which is suitable for moving target detection based on a non-coherent radar image, and broadens the time coverage of background model samples.

The real-time update method of the background model of the non-coherent radar image of the present invention comprises the following steps:

Step 1, establish a background model for each pixel in the radar image;

The gray value corresponding to each pixel (x, y) in the radar image is represented by v(x, y), and v _i represents the i-th background model sample, and the background model M(x ,y) is expressed as a set of N background model samples extracted from the previous N frames of images:

M(x,y)={v ₁ ,v ₂ ,...,v _N } (1)

The N background model samples in the set of formula (1) are initially obtained by the following method: For the i-th (1≤i≤N) frame image in the radar image sequence, from the 8-connected neighborhood N of the pixel (x, y) Extract the i-th background model sample v _i from _G (x, y):

在邻域N_G(x,y)中随机选择，

表示坐标处的灰度值。where the coordinates

randomly selected in the neighborhood N _G (x,y),

Indicates coordinates gray value at .

Step 2, read a frame of radar image, and classify new samples of each pixel;

According to the background model M(x,y) of the pixel (x,y), classify the new sample v(x,y) of the pixel in the current frame image, specifically: take v(x,y) as the center, and R is Radius, establish a spherical domain S _R (v(x,y)), if the number of samples in the intersection of the spherical domain and the background model sample set of M(x,y) is not less than the threshold # _min , then the new sample v( x, y) is judged as the background model sample, otherwise the new sample is judged as the foreground target. The intersection of the spherical domain and the background model sample set is expressed as:

#{S _R (v(x,y))∩{v ₁ ,v ₂ ,...,v _N }} (3)

Step 3, for a new sample determined to be a background model, update the background model of the corresponding pixel;

For a new sample v(x, y) judged to be a background model, a random decimal θ between 0 and 1 is generated. If θ is greater than the threshold T, the background model of the pixel (x, y) is updated, otherwise no update is performed.

When updating the background model, generate a random integer i between 0 and N, and replace the ith sample v _i in the background model with a new sample v(x, y).

Step 4, for the new sample determined as the background model, update the neighborhood pixel samples;

For the new sample v(x, y) judged as the background model, generate a random decimal θ between 0 and 1, if θ is greater than the threshold T, update the pixel background model in the 8-connected neighborhood of the pixel (x, y), Otherwise no update is performed.

When updating the pixel background model in the neighborhood, randomly select the pixel coordinates (x _NG , y _NG ) in the 8-connected neighborhood, generate a random integer i between 0 and N, and replace the pixel ( x _NG ,y _NG ) the i-th sample v _i in the background model.

Compared with the prior art, the real-time update method of the background model of the non-coherent radar image provided by the present invention prolongs the time for each sample to remain in the background model set while keeping the number of background model samples unchanged, broadening the The time coverage of the background model samples is improved.

Description of drawings

Fig. 1 is a schematic flow chart of a method for updating a background model of a non-coherent radar image in real time according to the present invention;

Fig. 2 is a frame of original non-coherent radar image and a certain pixel position according to the embodiment of the present invention.

Detailed ways

The following illustrates and describes the real-time update method of the background model proposed by the present invention in combination with the processing results of a non-coherent radar image sequence in the accompanying drawings.

The non-coherent radar background contains a large number of stationary objects, most of which are non-rigid targets (woods, grassland, water surface, etc.), the echo intensity fluctuates greatly, and the clutter interference at the edge of the background is strong, which makes it difficult to detect low-altitude small targets. The method of the present invention randomly sets the update cycle of the background model, and randomly replaces the samples in the background model, prolonging the retention time of each sample in the background model. The method for updating the background model in real time includes specific steps as follows:

Step 1, background model initialization; establish a background model for each pixel in the radar image.

First, the gray value corresponding to each pixel (x, y) in the radar image is represented by v(x, y), and each background model sample v _i is calibrated with the index i, i is an integer and 1≤i≤N, The background model M(x, y) of each pixel (x, y) consists of N background model samples extracted from the previous multi-frame images to form a background model sample set:

M(x,y)={v ₁ ,v ₂ ,...,v _N } (1)

For the first frame image in the radar image sequence (time t=0), the background model sample can be extracted from the 8-connected neighborhood _NG (x, y) of the pixel (x, y), denoted as M ⁰ (x, y):

可在邻域N_G(x,y)中随机选择，某个像素的灰度值

可能被多次选择，也可能从不被选择。v₁初始为M⁰(x,y)。Coordinates in the above formula

Can be randomly selected in the neighborhood N _G (x,y), the gray value of a pixel

May be selected multiple times, or may never be selected. v ₁ is initially M ⁰ (x,y).

最终得到如式(1)所示的像素(x,y)的背景模型。Continue to read N-1 frames of radar images. In each frame of image, the same method as shown in formula (2) is used to obtain the background model sample value of pixel (x, y) in each frame of image. For the i-th (1≤i≤N) frame image in the radar image sequence, the i-th background model sample v _i is extracted from the 8-connected neighborhood N _G (x, y) of the pixel (x, y):

Finally, the background model of the pixel (x, y) shown in formula (1) is obtained.

A frame of original non-coherent radar image is shown in Figure 2. The size of the image is 456×456. The coordinate origin is in the upper left corner of the image. The X-axis is horizontal to the right and the Y-axis is vertically downward. The pixel coordinates at the "*" position in the figure are is (138,360). The background model sample set of this pixel includes 10 samples in total, that is, N=10, each sample is the gray value of the pixel, and the background model is as follows:

M(138,360)＝{10,10,12,24,22,11,12,34,12,26} (3)

Step 2, read a frame of radar image, and classify the new samples of each pixel.

According to the current background model M(x,y) of the pixel (x,y), classify the new sample v(x,y) of the pixel (x,y) in the current frame image, and judge whether the new sample belongs to the background model sample set . With v(x,y) as the center and R as the radius, establish a spherical domain S _R (v(x,y)), where R is an integer. If the number of samples in the intersection of the spherical domain and the background model sample set is not less than the threshold # _min , then judge the new sample v(x,y) of the pixel as the background model sample, and proceed to steps 3 and 4; otherwise, judge The new sample v(x,y) of this pixel is not a background model sample, but a foreground target, and the following steps 3 and 4 are not performed. The intersection of the spherical domain and the background model sample set is expressed as:

#{S _R (v(x,y))∩{v ₁ ,v ₂ ,...,v _N }} (4)

In the embodiment of the present invention, it is assumed that the pixel gray value of the coordinate at the position "*" of the current frame image is v(138,360)=15, with the new sample as the center and R=5 as the radius, a spherical domain S _R=5 ( v(138,360)), threshold # _min =5, the intersection of the spherical domain and the background model sample set is

#{S _R＝5 (v(138,360))∩M(138,360)}＝{10,10,12,11,12,12} (5)

Then there are 6 samples in the intersection, which are greater than the threshold # _min , so this sample is a background sample.

Step 3, for a new sample determined to be a background model, update the background model of the corresponding pixel. In the update, the random setting of the update period and the update sample is realized.

For the new sample v(x, y) determined as the background model, it is determined whether to use it for background model update by generating a random decimal number θ between 0 and 1. If θ is larger than the threshold T, the background model of the pixel (x, y) is updated, otherwise no update is performed. The threshold T is set by the user and is data between 0 and 1.

When updating the background model, a random method is also used to select the update samples. Generate a random integer i between 0 and N, and replace the i-th sample v _i in the background model with v(x,y).

In the embodiment of the present invention, the threshold T=0.5 is set, and the generated random decimal is θ=0.68, then the background model is updated, and a certain sample in the original background model is replaced with a new sample v(138,360)=15. Generate a random integer i=6, then replace the sixth sample of the background model, the updated background model is

M(138,360)＝{10,10,12,24,22,15,12,34,12,26} (6)

Step 4, for the new sample judged as the background model, update the neighborhood pixel samples.

For a new sample determined as a background model, it can also be used to update the background model of the 8-connected neighborhood of the pixel (x, y).

The setting of the update period is the same as step 3, and the method of generating random decimals between 0 and 1 is used to judge whether to update. In the embodiment of the present invention, assuming that the generated random decimal is θ=0.52, which is greater than the threshold T=0.5, the neighborhood pixel background model is updated with a new sample v(138,360)=15.

Also use a random method to select the neighborhood pixel positions and replacement samples that need to be updated. Let (x _NG , y _NG ) be the pixel coordinates in the 8-connected neighborhood, including (x-1, y-1), (x-1, y), (x-1, y+1), (x, y -1), (x,y+1), (x+1,y-1), (x+1,y) and (x+1,y+1). After randomly determining the neighborhood coordinates (x _NG , y _NG ), the same as step 3, generate a random integer i between 0 and N, and replace the background model of the pixel (x _NG , y _NG ) with v(x, y) The i-th sample v _i in .

In the embodiment of the present invention, it is assumed that the neighborhood pixel coordinates (x _NG , y _NG )=(138,361) to be updated are randomly generated, and the original background model of the pixel is:

M(138,361)＝{15,12,14,26,26,17,22,28,22,28} (7)

Generate a random integer i=9, then replace the ninth sample of the pixel background model, the updated background model is

M(138,361)＝{15,12,14,26,26,17,22,28,15,28} (8)

After the background model is updated by using the current frame of radar image, the next frame of radar image can be read, and then step 2 is continued.

Claims (1)

1. a background model real time updating method for non-coherent radar image, is characterized in that, comprises the steps:

Step 1, for the each pixel in radar image is set up background model;

In radar image, gray-scale value corresponding to each pixel (x, y) represented by v (x, y), uses v _irepresent i background model sample, the background model M (x, y) of each pixel (x, y) is expressed as by N the background model sample set extracting in N two field picture before:

M(x,y)＝{v ₁,v ₂,...,v _N} (1)

N background model sample in formula (1) set, initially obtains: for i (1≤i≤N) two field picture in radar image sequence, be communicated with neighborhood N from 8 of pixel (x, y) by the following method _gin (x, y), extract i background model sample v _i:

Wherein, coordinate

at neighborhood N _gin (x, y), select at random,

denotation coordination the gray-scale value at place;

Step 2, reads a frame radar image, to the new samples classification of each pixel;

For pixel (x, y), according to corresponding background model M (x, y), the new samples v (x, y) of pixel in current frame image is classified, specifically: centered by v (x, y), R is radius, set up spherical territory S _r(v (x, y)), if the sample number in the common factor of the background model sample set of this spherical territory and M (x, y) is not less than threshold value # _min, new samples v (x, y) is judged as to background model sample, otherwise new samples is foreground target;

Step 3, for the new samples that is judged to be background model, upgrades the background model of respective pixel;

For the new samples v (x, y) that is judged to be background model, between generation 0～1, random decimal θ, if θ is greater than threshold value T, upgrades pixel (x, y) background model, otherwise does not upgrade;

In the time carrying out background model renewal, generate the random integers i between 0～N, and replace i sample v in current background model with new samples v (x, y) _i;

Step 4, for the new samples that is judged to be background model, upgrades the background model sample of pixel in neighborhood;

For the new samples v (x, y) that is judged to be background model, random decimal θ between generation 0～1, if θ is greater than threshold value T, is communicated with pixel background model in neighborhoods to 8 of pixel (x, y) and upgrades, otherwise do not upgrade;

When carrying out in neighborhood pixel background model while upgrading, select at random the pixel coordinate (x in 8 connection neighborhoods _nG, y _nG), generate the random integers i between 0～N, with new samples v (x, y) replacement pixel (x _nG, y _nG) background model in i sample v _i.

Images