CN103033166A - Target ranging method based on synthetic aperture focused images - Google Patents

Abstract

Translated fromChinese

本发明公开了一种基于合成孔径聚焦图像的目标测距方法。本发明包括：步骤1、利用小孔成像模型摄像机获取与目标视线垂直的等间隔线阵机位图像序列，线阵与目标视线交点处机位的图像作为基准图像；步骤2、将可测距离范围分成多个距离段，获取各距离段所对应的像差校正叠加图像；每一个距离段对应一幅像差校正叠加图像；步骤3、计算基准图像中每个像素的邻域与每一幅像差校正叠加图像中相应区域的相似度，并选取相似度随像差校正叠加图像变化的范围大于一预设阈值的像素作为可测距像素；步骤4、相似度最大的像差校正叠加图像所对应的距离段即为该可测距像素对应目标点所处的距离段。本发明具有实现成本低、抗干扰能力强、算法简单等优点。

The invention discloses a target ranging method based on a synthetic aperture focusing image. The present invention comprises: step 1, using a pinhole imaging model camera to acquire an image sequence of equally spaced line array positions perpendicular to the target line of sight, and taking the image of the position at the intersection of the line array and the target line of sight as a reference image; step 2, taking the measurable distance The range is divided into multiple distance segments, and the aberration-corrected superimposed images corresponding to each distance segment are obtained; each distance segment corresponds to an aberration-corrected superimposed image; step 3, calculating the neighborhood of each pixel in the reference image and each The similarity of the corresponding area in the aberration-corrected superimposed image, and select the pixels whose similarity varies with the aberration-corrected superimposed image and the range greater than a preset threshold as the distance-measurable pixels; step 4, the aberration-corrected superimposed image with the largest similarity The corresponding distance segment is the distance segment where the distance-measurable pixel corresponds to the target point. The invention has the advantages of low implementation cost, strong anti-interference ability, simple algorithm and the like.

Description

A kind of target ranging method based on the synthetic aperture focusing image

Technical field

The present invention relates to a kind of distance-finding method, relate in particular to a kind of target ranging method based on the synthetic aperture focusing image, belong to technical field of mapping.

Background technology

The method of measurement target distance has multiple, as based on the some telemetry of laser pulse mistiming, based on the degree of depth of Extent Modulate Wave phase differential obtain, geometrical optics focusing, Moire topologies, holographic interferometry, Fresnel diffractive technology and structured light method etc.Ultrasonicly also be commonly used to obtain depth information.Computer stereo vision technique is used widely in recent years in a lot of fields.Basic skills is to go to observe Same Scene from two or more viewpoints, the one group image of acquisition under different visual angles, obtain the parallax between respective pixel in the different images, then measure the depth information of target in the scene by triangulation calculation, it need to determine the corresponding point in binocular or the many orders image, and this is a very hard problem.When the space three-dimensional scene is projected as two dimensional image, some useful informations are owing to projection has been lost, same scenery can be very different in the image under the different points of view, blocked or the impact of shade, some points of scenery might not appear in all images, and the many changing factors in the scene, such as illumination condition, noise, the distortion of scene geometry, surface physical characteristic and camera properties etc., all be aggregated in the single gradation of image value, only determine that by this gray-scale value above factors is very difficult, this problem also is not well solved so far.Increase base length and can improve the depth survey precision, but can increase the difference between image simultaneously, increase the degree of difficulty of coupling.

Multimachine position pinhole imaging system can synthesize the large aperture image, and is consistent with single-lens imaging.There are document [Kusumoto, N. in Japan; Hiura, S.; Sato, K.Uncalibrated Synthetic Aperture for Defocus Control IEEEConference on Computer Vision and Pattern Recognition, 2009.CVPR 2009.P:2252-2259] studied and utilized the synthetic aperture method that the pin-hole imaging photo is carried out art processing, make it to produce non-main body defocus effect, with outstanding case of visual emphasis.The synthetic aperture radar image-forming technology is ripe at present.Monocular is poly-/and the development of range of defocusing method is comparatively ripe, use a plurality of video cameras can obtain monocular large aperture effect according to the synthetic aperture principle, therefore can use for reference abundant the gathering/the range of defocusing algorithm of monocular imaging, so that the synthetic aperture imaging range finding has higher development foundation.Because it is very low to meet the digital equipment cost of pin-hole imaging characteristics, imaging results is convenient to digitized processing, and the development prospect of synthetic aperture imaging is wide.

Summary of the invention

Technical matters to be solved by this invention is to overcome the deficiency of existing ranging technology, and a kind of target ranging method based on the synthetic aperture focusing image is provided, and has the advantages such as the cost of realization is low, antijamming capability is strong, algorithm is simple.

Target ranging method based on the synthetic aperture focusing image of the present invention may further comprise the steps:

Step 1, utilize the pinhole imaging system model camera to obtain the uniformly-spaced linear array seat in the plane image sequence vertical with line of sight, the image of linear array and seat in the plane, line of sight intersection point place is as benchmark image;

Step 2, can find range is divided into a plurality of distance segment from scope, for each distance segment, calculate respectively first the aberration between each width of cloth image and benchmark image in the described image sequence, then each width of cloth image in the image sequence is carried out superposeing behind the aberration correction, obtain the corresponding aberration correction superimposed image of this distance segment; The corresponding width of cloth aberration correction superimposed image of each distance segment;

Step 3, calculate each pixel in the benchmark image neighborhood and each width of cloth aberration correction superimposed image in the similarity of respective regions, and choose scope that similarity changes with the aberration correction superimposed image greater than the pixel conduct of the predetermined threshold value pixel of can finding range;

Step 4, for each the found range pixel in the benchmark image, select the aberration correction superimposed image of similarity maximum of the neighborhood of respective regions and this pixel of can finding range, the corresponding distance segment of this aberration correction superimposed image is this residing distance segment of the corresponding impact point of pixel of can finding range.

In the technique scheme, described uniformly-spaced linear array seat in the plane image sequence can utilize a plurality of video cameras of uniformly-spaced arranging to take simultaneously simultaneously or not and obtain, also can utilize same pinhole imaging system model camera uniformly-spaced follow shot obtain, especially the latter for take photo by plane the range finding application have more significance.

Compared to existing technology, the present invention has following beneficial effect:

1) be embodied as the flexible movement (with the flexible hypertonic of aperture) of image planes with the real time signal processing method, realize that monocular focuses on measuring distance, avoided the mechanical action of optical lens focal imaging, measuring speed is fast, and is easy-to-use;

2) compare with the manufacturing process of optics high-aperture lenses complexity, it is easier to form same large synthetic aperture, and large synthetic aperture focusing distance accuracy is high, can be widely used in the fields such as production, scientific research, military affairs, especially can be applicable to aerial image sequence, survey and draw out the ground scene 3-D view;

3) passive mapping have hidden, without the advantage of Contamination of Electromagnetic Wave, the mapping cost low;

4) if initiatively to target area projective textures pattern, can make target area image information entropy reasonable layout, but the point distance measurement distribution suits the requirements;

5) because linear array adjacent camera baseline is very short, the shielded image that mild viewpoint difference causes changes, the image sequence synthesized image has obvious inhibition ability to differences such as single image illumination, DE Camera Shake, noises, and this method does not need to carry out images match, and algorithm is stable.

Description of drawings

Fig. 1 a～Fig. 1 d is the principle schematic of synthetic aperture focusing imaging, wherein, and Fig. 1 a display-object point p₁Focus on the imaging surface through the large aperture optical lens; Fig. 1 b shows has a sub-aperture apart from photocentre line r place, through this sub-aperture imaging, and p₁Still focus on the picture center, and the impact point p of another distance₂There is aberration ε in imaging apart from the picture center, it and r linear positive correlation; Fig. 1 c, Fig. 1 d show full aperture imaging p₁Be a bit, and p₂Be imaged as a circle, it is comprised of each sub-aperture imaging;

Fig. 2 is linear array of the present invention seat in the plane synoptic diagram;

Fig. 3 the present invention is based on the schematic flow sheet of the target ranging method of synthetic aperture focusing image;

Fig. 4 is linear array seat in the plane image sequence example uniformly-spaced;

Fig. 5 a, Fig. 5 b are respectively more closely and the corresponding aberration correction superimposed image of two distance segment far away;

Fig. 6 is the information entropy distribution binary image of benchmark image;

Fig. 7 is synthetic aperture target depth image.

Embodiment

Below in conjunction with accompanying drawing technical scheme of the present invention is elaborated:

Fully understand technical scheme of the present invention for the ease of the public, at first the ultimate principle of Technologies Against Synthetic Aperture focal imaging is introduced:

During the high-aperture lenses imaging, the incident ray by each position of camera lens is after the camera lens refraction, and target light is assembled again, becomes two-dimensional image at imaging surface.If incident ray is pressed the sub-aperture separate imaging that passes through, with the signal processing method stack, just obtain the synthetic aperture picture again, imaging results is consistent with the high-aperture lenses imaging.Fig. 1 a display-object point p₁Focus on the imaging surface through the large aperture optical lens; Fig. 1 b shows has a sub-aperture apart from photocentre line r place, through this sub-aperture imaging, and p₁Still focus on the picture center, and the impact point p of another distance₂There is aberration ε in imaging apart from the picture center, it and r linear positive correlation.Fig. 1 c, Fig. 1 d show full aperture imaging p₁Be a bit, and p₂Be imaged as a circle, it is comprised of each sub-aperture imaging, therefore we can emulation high-aperture lenses imaging process, each sub-aperture image is superposeed by certain rule, obtain the synthetic aperture picture, it has the feature of high-aperture lenses imaging, and the method that can use the monocular range finding is measured the distance of synthetic aperture target.Form the synthetic aperture picture, and do not require that all sub-aperture images all participate in calculating in Fig. 1 d macropore collar, only need the parton aperture image to participate in superposition calculation, just can form the synthetic aperture effect.

Thinking of the present invention is exactly to utilize the principle of synthetic aperture focusing imaging to carry out object ranging, specifically as shown in Figure 3, may further comprise the steps:

Step 1, utilize the pinhole imaging system model camera to obtain the uniformly-spaced linear array seat in the plane image sequence vertical with line of sight, the image of linear array and seat in the plane, line of sight intersection point place is as benchmark image.

Above-mentioned image sequence can utilize uniformly-spaced a plurality of video cameras of arranging to take simultaneously simultaneously or and obtain, also can utilize same pinhole imaging system model camera uniformly-spaced follow shot obtain (for example aerial image sequence).Describe as an example of video camera linear array shown in Figure 2 example in this embodiment, each video camera imaging meets the pinhole imaging system model in the linear array, and each video camera can be regarded the sub-aperture of linear array of high-aperture lenses as like this.As shown in Figure 2, the 2N+1 in this video camera linear array identical video camera C_nParallel placement, n=-N ..., 0 ... N, C₀Be fiducial cameras, all video cameras evenly are fixed on the cross bar that length is 2NB, and the video camera spacing is B, fiducial cameras C₀At the cross bar center.The lens direction of video camera is vertical with cross bar, all video camera axis coplanar.So, video camera C_nWith fiducial cameras C₀Baseline be nB.Utilize in this video camera linear array video camera simultaneously or simultaneously target is not taken, can obtain comprising the image sequence of 2N+1 width of cloth image.Fig. 4 is an example of linear array seat in the plane image sequence uniformly-spaced, and video camera for shooting is the aperture camera lens of the super depth of field, and image sequence is not simultaneously collection, and each image irradiation is widely different.Synthetic aperture focusing stereoscopic vision requires to obtain a plurality of images of the different seats in the plane of same target, and illumination, video camera difference, noise etc. are had stronger adaptive faculty.Image sequence can from video camera array, also can be that same camera shifting timesharing is taken.In fact aerial image sequence is processed, and the mobile camera image sequence processing also can adopt the synthetic aperture principle to obtain depth information.

Step 2, can find range is divided into a plurality of distance segment from scope, for each distance segment, calculate respectively first the aberration between each width of cloth image and benchmark image in the described image sequence, then each width of cloth image in the image sequence is carried out superposeing behind the aberration correction, obtain the corresponding aberration correction superimposed image of this distance segment; The corresponding width of cloth aberration correction superimposed image of each distance segment.

Still the video camera linear array in Fig. 2 is as example, and the initial point of each camera coordinate system is the photocentre of camera lens separately, and Z axis is by photocentre direction field sight spot direction, and perpendicular to the plane of delineation, along base direction, perpendicular to other diaxon, this three satisfies the right-hand rule to Y-axis to X-axis by photocentre by photocentre.x_nBe impact point P horizontal ordinate in each image coordinate system, z be P to the distance of video camera principal plane, nB is base length, f is focal length.Can draw video camera C by geometric relationship_nWith fiducial cameras C₀Aberration be:

L wherein_cBe the target surface pixel distance.Aberration between each image and the benchmark image in (distance segment that different values is corresponding different) image sequence in the time of can obtaining z and get different value according to following formula.For each distance segment, according to the aberration that calculates each width of cloth image in the sequence is carried out aberration correction, and the 2N+1 width of cloth image behind the aberration correction superposeed, then obtain to aberration correction superimposed image that should distance segment the corresponding width of cloth aberration correction superimposed image of each distance segment.If fiducial cameras C₀Image for being designated as s₀(x, y), then superposed signal is:

$s_{\mathrm{\Σ}}(x,y)=\underset{n=-N}{\overset{N}{\mathrm{\Σ}}}{s}_0(x-\frac{\mathrm{nBf}}{L_{c}z},y)=\underset{n=-N}{\overset{N}{\mathrm{\Σ}}}{s}_0(\frac{\mathrm{Bf}}{L_{c}z}(\frac{L_{c}z}{\mathrm{Bf}}x-n),y)h\left(n\right)=s_0(x,y)*h\left(\frac{L_{c}z}{\mathrm{Bf}}x\right)$

Wherein$h\left(n\right)=\left\{\begin{array}{cc}1& -N\≤n\≤N\\ 0& \mathrm{else}\end{array}\right.$For synthetic aperture pixel diffusion kernel, different from optical aperture, synthetic aperture nuclear proliferation scope can be very large, and diffusion way can disperse, and diffusion peacekeeping direction also can artificially design.Displacement can be sub-pixel, can make the synthetic aperture image focu in any distance, and namely imaging surface can move arbitrarily.Directly all image sequence stack gained synthesized images are focused on the infinite point target.Video camera linear array length 2NB is length of synthetic aperture, gives up linear array two ends camera review and can be analogous to the aperture that has dwindled process shot.Because linear array adjacent camera baseline is very short, alleviated the shielded image variation that the viewpoint difference causes.Fig. 5 a and Fig. 5 b have shown two width of cloth image sequence synthesized images, and Fig. 5 a focuses on the close shot place, and Fig. 5 b focuses on relative distant view place, because the synthetic aperture diameter is more much bigger than optical lens, so small target range difference also can cause significantly poly-/defocusing effect.The image sequence synthesized image has obvious inhibition ability to differences such as single image illumination, DE Camera Shake, noises.

Step 3, calculate each pixel in the benchmark image neighborhood and each width of cloth aberration correction superimposed image in the similarity of respective regions, and choose scope that similarity changes with the aberration correction superimposed image greater than the pixel conduct of the predetermined threshold value pixel of can finding range.

The similarity of area image is defined as in the subregion:$R_z(x,y)=\frac{{\mathrm{\Σ}}_{x,y}{s}_{\mathrm{\Σ}}(x,y)s_0(x,y)}{{\mathrm{\Σ}}_{x,y}{s}_{\mathrm{\Σ}}(x,y){\mathrm{\Σ}}_{x,y}{s}_0(x,y)},$(x, y) is pixel coordinate in the subregion in the formula.For example, getting the length of side is the square slide block of 5 pixels, from left to right, sliding all over the view picture picture successively from top to bottom in picture as slide block center pixel neighborhood, calculates benchmark image and stacking diagram's similarity R in the slide block_zWhen (x, y), only need to revise ∑ corresponding to border variation pixel that slide block slides and causes_{X, y}s_∑(x, y) s₀(x, y), ∑_{X, y}s_∑(x, y) and ∑_{X, y}s₀(x, y) just change can so that the similarity counting yield is high.Change the z value with orderly interval and calculate R_z(x, y) repeats said process, obtains R corresponding to all z values (distance segment)_z(x, y).Like this, all there is the similarity value of a correspondence in each pixel in the benchmark image under each distance segment.

The target area image entropy can not be too little when using the focusing effect range finding, be that target area image will have certain quantity of information, otherwise can not produce the poly-difference that defocuses, Fig. 6 has shown benchmark image information entropy distribution binary image, wherein white portion represents that this area image has enough target information amount, can find range, and the null value district represents that this area image does not have enough target information amount, cannot find range.Yet, very complicated with the information entropy split image, can suffer from each image sampling condition difference.In fact by investigating R_z(x, y) just can distinguish poly-defocus difference and split image.Its R of image-region that entropy is little_z(x, y) is with z(or distance segment) variation is little, on the contrary then dynamic change scope is large.Give similarity R_z(x, y) is with z(or distance segment) the dynamic change scope establishes a threshold value, to each neighborhood of pixels in the benchmark image, similarity R_z(x, y) dynamic change scope (being the poor of similarity maximal value and minimum value) is only effective range finding greater than the object ranging of this threshold value, and this pixel is for finding range pixel, and the pixel that is less than or equal to this threshold value then should not be used to find range.

Step 4, for each the found range pixel in the benchmark image, select the aberration correction superimposed image of similarity maximum of the neighborhood of respective regions and this pixel of can finding range, the corresponding distance segment of this aberration correction superimposed image is this residing distance segment of the corresponding impact point of pixel of can finding range.

For the found range pixel in the benchmark image, its maximum similarity R_z(x, y) corresponding z value (distance segment) is the distance segment at the corresponding impact point of this pixel place.Distance according to the corresponding impact point of pixel of respectively can finding range that obtains can obtain synthetic aperture target depth image as shown in Figure 7, and brightness represents target depth among the figure, and the yardstick unit of the brightness in left side and distance collation table is rice.Utilize this figure in conjunction with Fig. 6, can get the degree of depth (distance) at target effective position.

The target ranging method that the present invention is based on the synthetic aperture focusing image only focuses on the specific range target, focuses on the echo signal concentration of energy, is subjected to other distance objective effect of signals little, focusing as concentration of energy in the zonule, be easy to distinguish with other distance objectives, algorithm is stable, is suitable for engineering and uses.By$x_0-x_n=\frac{\mathrm{nBf}}{{\mathrm{zL}}_c}$$\mathrm{dz}=\frac{z^2}{\mathrm{nBf}}d(x_0-x_n),$n=-N,…,0,…N。Minimum value wherein isThis shows that range resolution is directly proportional with square distance, and distance distance accuracy far away is poorer.D (x in the formula₀-x_n) be that image can be distinguished pixel displacement, jointly determined by algorithm and system hardware, usually pixel distance on the target surface as d (x₀-x_n) assessed value, it is subjected to the hardware advances horizontal constraints.As seen increase the length of B or select long focal length lens can improve resolution, but can bring finding range to change.It has been generally acknowledged that but the range finding maximum distance is subjected to the size restriction of target recognition image, the range finding minimum distance is tested apart from the dead band restriction, enter the range finding dead band after, have the partial line array camera to cannot see target, affect the figure image focu and finds range.The video camera distribution density, whether whether evenly distribute affects non-distance z place scene image by evenly level and smooth.The number of cameras increase means that calculated amount increases and systematic error increases.

The synthetic aperture focusing range finding need utilize texture and the marginal information of target image, if initiatively to target area projective textures pattern, can make target area image information entropy reasonable layout, point distance measurement suits the requirements.Because linear array adjacent camera baseline is very short, the shielded image that mild viewpoint difference causes changes.This algorithm does not need to carry out images match, directly focuses on target, and algorithm is extremely stable, can process in real time.

Claims (2)

1. the target ranging method based on the synthetic aperture focusing image is characterized in that, may further comprise the steps:

Step 1, utilize the pinhole imaging system model camera to obtain the uniformly-spaced linear array seat in the plane image sequence vertical with line of sight, the image of linear array and seat in the plane, line of sight intersection point place is as benchmark image;

Step 2, can find range is divided into a plurality of distance segment from scope, for each distance segment, calculate respectively first the aberration between each width of cloth image and benchmark image in the described image sequence, then each width of cloth image in the image sequence is carried out superposeing behind the aberration correction, obtain the corresponding aberration correction superimposed image of this distance segment; The corresponding width of cloth aberration correction superimposed image of each distance segment;

Step 3, calculate each pixel in the benchmark image neighborhood and each width of cloth aberration correction superimposed image in the similarity of respective regions, and choose scope that similarity changes with the aberration correction superimposed image greater than the pixel conduct of the predetermined threshold value pixel of can finding range;

Step 4, for each the found range pixel in the benchmark image, select the aberration correction superimposed image of similarity maximum of the neighborhood of respective regions and this pixel of can finding range, the corresponding distance segment of this aberration correction superimposed image is this residing distance segment of the corresponding impact point of pixel of can finding range.

2. as claimed in claim 1 based on the target ranging method of synthetic aperture focusing image, it is characterized in that, described uniformly-spaced linear array seat in the plane image sequence utilize same pinhole imaging system model camera uniformly-spaced follow shot obtain.

Images