CN101447083A - Beaconing-free vision measuring-technique for moving target based on time-space correlative characteristics - Google Patents

Abstract

The invention relates to a beaconing-free vision measuring-technique for a moving target based on time-space correlative characteristics, which belongs to the technical field of vision measuring, aims at overcoming shortages and shortcomings in the prior art and providing a moving target identifying and detecting technique which the beaconing-free vision measuring-technique, namely, a signal marker is not needed on a target motion body in advance. When in technique realization, the time-space correlative characteristics of motion pixels are considered at the same time, a background image difference method and an inter-frame image difference method are combined together to identify the moving target, so that the disadvantages thereof are overcome effectively, the self-adaptive ability is strong, the speed is high, the antinoise effect is good proved by experiments, and the moving target can be better detected from video images.

Description

Beaconing-free vision measuring-technique for moving target based on time-space correlative characteristics

Technical field

The invention belongs to technical field of visual measurement, relate to a kind of utilize time-space correlative characteristics exempt from beacon moving target recognition technology.

Background technology

Common moving target vision measurement technology has three kinds of basic skills: background image method of difference, frame-to-frame differences point-score and optical flow method.The detected moving target of background image method of difference position is accurate, and speed is fast, but in actual applications because two width of cloth images that compare were taken the photograph from the different moment, therefore is vulnerable to natural cause and other artificial factor such as illumination variation, wind.And in some cases, the partial pixel gray-scale value in the shared zone may be identical with the background pixel gray-scale value on the corresponding region in the reference picture on image for research object.Carried out the research of background reconstruction at present, in the hope of reducing dynamic scene for the influence of accurately cutting apart, but these methods also are not very perfect; The frame-to-frame differences point-score only detects the object of relative motion, because of the time interval between two width of cloth images shorter, difference image is subjected to the light variable effect little, detects effectively and stablizes.But these class methods have two defectives: (1) two interframe overlapped object part does not detect, and promptly only detects the part of object, forms the cavity in the middle of object; (2) the position out of true of detected object, its boundary rectangle is stretched on direction of motion, and detected target is bigger than real object; The advantage of optical flow method is to detect the object of self-movement, and does not need to know in advance any information of scene, and shortcoming is that operational formula complexity, calculated amount are big, is difficult to reach real-time requirement under the condition that does not have the special hardware support.

In the vision measurement of moving target, the detection and the identification of movable body is adopted usually the method for impact point feature identification.The specific signal thing promptly is installed on the target travel body in advance, or on objective body, is chosen specific monumented point in advance, realize identification and location the target travel body by image recognition to beacon (marker thing).The method of this employing beacon identification not only requires high to hardware environment, and often there be inefficacys in beacon, problem such as cover, come off, and final identification to moving target brings very big influence.

Summary of the invention

At the problems referred to above, the present invention proposes a kind of beaconing-free vision measuring-technique for moving target based on time-space correlative characteristics.Vision measurement technology proposed by the invention is based on time-space correlative characteristics, has promptly taken into full account the time and the spatial coherence of motion pixel.In technical scheme, the image sequence that order is relevant on the time is analyzed; Meanwhile, for the moving target on the different time points, also taken into full account its relevance on the image two-dimensional space.This method plays a good role for the effect of vision measurement.Be different from and utilize beacon to carry out the method for moving object detection usually, this technology is a kind of vision measurement technology of exempting from beacon.The beacon of exempting from that proposes among the present invention is meant: need not to install in advance any signal thing on moving target, also need not to choose in advance specific monumented point on moving target.Therefore, utilize the present invention under the situation that any beacon is not set, to carry out vision-based detection to moving target.

Technical scheme specifically is expressed as follows: the present invention is divided into two image-regions with two field picture, utilizes the background image method of difference, carries out the moving target rough detection at image-region one.Utilize the temporal correlation of target travel, when moving target enters image-region two fully, in image-region two these regional areas, carry out adjoining inter-frame difference computing, be partitioned into moving target.

The step of specifically taking is as follows:

1. at first obtain the image U that a width of cloth only contains fixed background (x, y);

2. from sequence image, obtain current frame image F _k(x, y);

With U (x, y) and F _k(x y) is divided into two image-region U by row respectively ¹(x, y), U ²(x, y),

4.U ¹(x, y) with

Difference obtains the image Δ ¹(x, y):

${\mathrm{\Δ}}^1(x,y)=|F_k^1(x,y)-U^1(x,y)|$

5. utilize criterion parameter P _aTo Δ ¹(x, y) further calculate:

6. computed image G ¹(average and standard deviation calculation formulas are as follows for x, average y) and standard deviation:

$\mathrm{\μ}=\frac{1}{m\×n}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}f(x,y)$

$\mathrm{\δ}={\left[\frac{1}{m\×n}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}{\left(f(x,y-\mathrm{\μ})\right)}^2\right]}^{\frac{1}{2}}$

Wherein m, n are respectively the line number and the columns of image, and (x y) is (x, y) gray-scale value at some place to f.

7. to average and standard deviation ranking operation and obtain evaluation index Ω, computing formula is as follows:

Ω＝k ₁μ+k ₂δ

Latch an other two field picture M _k(x, y), M _k(x y) also is divided into two image-regions

8. use

With

Difference obtains image G ²(x, y)

The present invention combines background image method of difference and inter frame image method of difference and carries out moving target identification, overcome the shortcoming of the two effectively, adaptive ability, speed are fast, the experiment proved that noise robustness is better, can detect moving target preferably from video image.

The present invention has following technical characterictics:

(1) background image and current frame image will be divided into two image-regions by row respectively;

(2) image that obtains of difference will utilize the criterion parameter, further the average and the standard deviation of computed image;

(3) obtain evaluation index by average and standard deviation ranking operation;

(4) introduce the criterion parameter to determine the appearance of moving target;

(5) choosing of time delay will guarantee that moving target can enter in the image-region fully.

Description of drawings

Fig. 1: detection algorithm process flow diagram

Fig. 2: the object in the high-speed motion mechanism

Fig. 3: T ₁Moment images acquired and recognition result

Fig. 4: T ₂Moment images acquired and recognition result

Embodiment

Below in conjunction with accompanying drawing and instantiation the present invention is described in further detail.Apply the present invention to detect and discern an object in the high-speed motion mechanism, as shown in Figure 2.When target travel, real-time images acquired, and by the method identification moving target.

The first step, when motion is static, the background image U of collection object (x, y).

(x y) needs regularly to upgrade U.Set a timer, when the time to and when not finding moving target, just with current frame image background image updating U (x, y).

In second step, motion begins high-speed motion, the object accompany movement in the mechanism.Gather current frame image F according to sampling parameter _k(x y), and deposits in the working storage.

The 3rd step, with the image U that obtains (x, y) and F _k(x y) is divided into two image-region U by row respectively ¹(x, y), U ²(x, y), $F_k^1(x,y),F_k^2(x,y).$

The 4th step is according to background image selected threshold TH _MaxAnd TH _Min, with image U ¹(x, y) with

Image after the binaryzation, binaryzation adopts 3 * 3 template filtering and noise reduction sound, and two width of cloth image individual elements are done calculus of differences and drawn the difference image Δ then ¹(x, y):

${\mathrm{\Δ}}^1(x,y)=|F_k^1(x,y)-U^1(x,y)|$

In the 5th step, utilize criterion parameter P _aTo Δ ¹(x, y) further calculate:

Criterion parameter P _aChoose according to parameters such as sampling rate, image resolution ratios and decide.Usually, in vision measurement, parameters such as the sampling rate of detection system, image resolution ratio are changeless.

The 6th step, computed image G ¹(average and standard deviation calculation formulas are as follows for x, average y) and standard deviation:

$\mathrm{\μ}=\frac{1}{m\×n}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}f(x,y)$

$\mathrm{\δ}={\left[\frac{1}{m\×n}\underset{x=1}{\overset{m}{\mathrm{\Σ}}}\underset{y=1}{\overset{n}{\mathrm{\Σ}}}{\left(f(x,y-\mathrm{\μ})\right)}^2\right]}^{\frac{1}{2}}$

Wherein m, n are respectively the line number and the columns of image, and (x y) is (x, y) gray-scale value at some place to f.

In the 7th step, to average and standard deviation ranking operation and obtain evaluation index Ω, computing formula is as follows:

Ω＝k ₁μ+k ₂δ

When Ω greater than P _bThe time, define moving target and occur.Occur if find moving target, behind delay a period of time T, latch an other two field picture M _k(x, y), M _k(x y) also is divided into two image-regions

The selection of T will guarantee that moving target enters fully behind Ts In the image-region.

In the 8th step, use

With

Difference obtains image G ²(x, y), G ²(x y) has characterized the variation of moving target in the image, to G ²(x, y) further processing can obtain the position and the profile of moving target.

The present invention organically combines background image method of difference and inter frame image method of difference, at first in image-region one, moving target is carried out Rough Inspection with the background image method of difference, after target occurs and enters image-region two fully, be partitioned into moving target with adjoining inter frame image method of difference.The division of image-region is decided according to the speed of moving target, guarantee that moving target can not enter image-region two on the first appearance.(x y) needs regularly to upgrade the fixed background U that uses in this method.Set a timer, when the time to and when not finding moving target, just use the current frame image background image updating.

Claims (10)

1 beaconing-free vision measuring-technique for moving target based on time-space correlative characteristics comprises the steps: at first two field picture to be divided into two image-regions, utilizes the background image method of difference then, carries out the moving target rough detection at image-region one.Utilize the temporal correlation of target travel again, when moving target enters image-region two fully, in image-region two these regional areas, carry out adjoining inter-frame difference computing, be partitioned into moving target.

2 by the described method of claim 1, and the beacon of exempting from that proposes among the present invention is meant: need not to install in advance any signal thing on moving target, also need not to choose in advance specific monumented point on moving target.

3 by the described method of claim 1, it is characterized in that obtaining background image U (x, y) after, constantly from sequence image, obtain current frame image F _k(x, y), and with U (x, y) and F _k(x y) is divided into two image-regions by row respectively.

4 by the described method of claim 1, it is characterized in that utilizing the background image method of difference, carries out computing at a pair of current frame image of image-region and background image, obtains difference image:

${\mathrm{\Δ}}^1(x,y)=|F_k^1(x,y)-U^1(x,y)|$

Wherein , U ¹(x y) belongs to image-region one scope.

5 by the described method of claim 1, it is characterized in that introducing criterion parameter P _aTo determine the appearance of moving target.

6 by the described method of claim 4, it is characterized in that utilizing P _aTo the difference image Δ ¹(x y) carries out obtaining evaluation index Ω after the conversion.

7 by the described method of claim 5, it is characterized in that computed image G ¹(x, average y) and standard deviation, wherein:

8 by the described method of claim 5, it is characterized in that average and standard deviation ranking operation and obtains evaluation index Ω, and computing formula is as follows:

Ω＝k ₁μ+k ₂δ。

9 by the described method of claim 1, after it is characterized in that determining the moving target appearance and postponing a period of time T, gathers a two field picture M _k(x y), calculates

With

Difference obtain image G ²(x, y).

10 by the described method of claim 8, it is characterized in that difference image G ²(x, computing formula y) is as follows:

Images