CN103780897B - Depth map acquisition methods and the device of two visual point images - Google Patents
Depth map acquisition methods and the device of two visual point images Download PDFInfo
- Publication number
- CN103780897B CN103780897B CN201410064382.4A CN201410064382A CN103780897B CN 103780897 B CN103780897 B CN 103780897B CN 201410064382 A CN201410064382 A CN 201410064382A CN 103780897 B CN103780897 B CN 103780897B
- Authority
- CN
- China
- Prior art keywords
- visual point
- subimage
- point image
- depth map
- pixel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Abstract
The present invention relates to technical field of image processing, relate to depth map acquisition methods and the device of two visual point images. The method comprises carries out resolution ratio normalization by two visual point images of rgb format; Two visual point images after resolution ratio normalization are converted to yuv format, and extract acquisition Y and divide spirogram, wherein Y divides spirogram to comprise left visual point image and right visual point image; Left visual point image and right visual point image are carried out to cutting apart of same format, to form multiple subimages pair in left visual point image and right visual point image; Parallel to multiple subimages to carrying out motion search, obtain left depth map and right depth map; Respectively the depth value in left depth map and right depth map is mapped in 0～255 scope, obtains the depth map of two visual point images. Depth map acquisition methods and the device of two visual point images of the embodiment of the present invention, data processing amount is little, and data processing real-time is high, is easy to realize the actual demand that more can meet user and obtain the depth map of two visual point images.
Description
Technical field
The present invention relates to technical field of image processing, in particular to two visual point imagesDepth map acquisition methods and device.
Background technology
China's 3D industry development is swift and violent, and the more entirety of 3D industry of opening of three-dimensional channel is sent outExhibition brings major opportunity. In 3D industry development process, the technology that 3D terminal shows rapidly moreNewly, but the development of 3D programme content lag far behind. Along with digital image acquisition apparatus skillThe progress of art, people have obtained the media materials such as a large amount of two-dimensional videos, image, photo.Utilizing existing huge two-dimentional resource to meet in the process of 3D industry development demand, needAdopt the technology such as the recovery of 3D information and scene rebuilding to carry out the making of 3D programme content. GrindStudy carefully and show, in three dimensions, same object has the position in horizontal direction in the time of right and left eyes imagingMove, this is called as " parallax ". Research to mankind's physiology stereoscopic vision key element points out, due toThe existence of parallax, the about mankind can produce third dimension when Same Scene soon. Object in sceneBe that depth information is to produce the main cause of parallax apart from the distance of camera position, the two itBetween exist corresponding relation. The original steric information of reduction two dimensional image is (flat perpendicular to imageThe depth information of the third dimension degree of face) be the important content that two dimensional image converts 3-D view to.In same actual life, two visual point images are on the increase, but the stereoscopic display of two visual point imagesEffect is good not as good as the display effect of the multi-view image of 8 viewpoints and so on far away, has therefore proposedTwo visual point images are converted to the actual demand of multi-view image. Two visual point images are being changedBecome in the process of multi-view image, two visual point images are obtained in an important previous workDepth map.
The method of obtaining at present the depth map of two visual point images mainly comprises: (1) is based on how muchThe method of perspective, is specially the geometrical-restriction relation utilizing in perspective imaging and determines destination objectSpace three-dimensional attitude and the three-dimensional depth information of whole scene, it is mainly applicable to city and buildsBuild thing etc. and contain the comparatively scene of regular shape object; (2) method based on geometric optics,Be specially focusing or defocusness method: focusing is in the adjustable situation of focal length, makes in imageImpact point vernier focusing, then tries to achieve this distance with respect to camera according to lens imaging principleFrom, the method expensive hardware, difficult realization, error will definitely not brought in vernier focusing location;Defocusness method according to each point in fog-level computed image with respect to the distance of camera, but how accurateReally setting up and defocusing model is Major Difficulties; (3) adopt machine learning and Bayesian inference methodCarry out haplopia estimation of Depth and scene rebuilding, this kind of method utilized multiple Depth cue and imageFeature, obtains the depth map of testing image by the method for training and learn, but this sideMethod need to gather training data, set up Sample Storehouse, and the performance of calculating also needs further to be improved.
Find out thus, the operand that obtains two visual point images in prior art is larger, and is correlated withIn technology, the single threads that adopt carry out computing more, are difficult to obtain in time depth map, do not meet and useFamily obtains the actual demand of depth map.
Summary of the invention
The object of the present invention is to provide depth map acquisition methods and the device of two visual point images,To solve the above problems.
Provide in an embodiment of the present invention the depth map acquisition methods of two visual point images, bagDraw together: two visual point images of RGB rgb format are carried out to resolution ratio normalization; By resolution ratioDescribed two visual point images after normalization are converted to yuv format, and extract acquisition brightness letterNumber Y divides spirogram, and wherein said Y divides spirogram to comprise left visual point image and right visual point image; WillDescribed left visual point image and described right visual point image carry out cutting apart of same format, with describedIn left visual point image and described right visual point image, form position multiple subimages of correspondence mutuallyRight; For each subimage pair, using the subimage of described left visual point image as for working asThe current subimage of front operation; Make center and the described current son of default search patternThe pixel that meets preset rules of image overlaps; Taking the subimage of described right visual point image asSearch subimage; In the same position of described search subimage, described search pattern is set; ?In described search subimage, move described search pattern along the width of described search subimageCenter, all calculate described search when the center of mobile described search pattern at every turnSubimage be positioned at described search pattern scope pixel absolute error and, when described absolutelyStop the motion search of described search pattern to error with while being less than predetermined threshold value, and record instituteState the distance that move search pattern center; All subimages to all complete motion search itAfter, utilize the distance of record to obtain left depth map; For each subimage pair, with describedThe subimage of left visual point image is as search subimage, according to the side of obtaining of described left depth mapFormula obtains right depth map; Respectively by the depth value in described left depth map and described right depth mapBe mapped in 0～255 scope, obtain the depth map of two visual point images;
Wherein, described absolute error and the following method of employing obtain:
Set up xoy coordinate system, in described xoy coordinate system, with described left visual point image pairThe function of answering is designated as L (x, y), and the function corresponding with described right visual point image is designated as R (x, y);
When each center of moving described search pattern, calculating described search subimage is positioned atThe absolute error of the pixel within the scope of described search pattern and computing formula be,
f(k)＝Σ(x∈1-9)Σ(y∈1-5)|M(x,y)*L(x,y)-M(x+k,y)*R(x,y)|；
Wherein M(x,y)For width is 9 pixels, be highly the search pattern of 5 pixels, f (k) isDifference mean value, the distance that move the center that k is described search pattern;
Wherein, the described pixel that meets preset rules comprises: the pixel that pixel value is non-vanishingPoint or pixel coordinate value meet the pixel of x%2==0&&y%2==0.
Preferably, described extraction obtains Y and divides spirogram, comprising: described Y divides spirogram by YComponent value composition, described Y component value is obtained by conversion in two visual point images of rgb format,Wherein the transfer algorithm of Y component value is: Y=R*I+G*J+B*K;
Wherein, Y represents Y component value, and R, G and B represent that respectively two of GRB form looksThe red component R value of a pixel, green component G value and blue component B value in dot image,I, J and K represent respectively red component R value, green component G value and blue component B valueWeighted value, and meet: I+J+K=1 and I:J:K ≈ 1:4.3:2.7.
Preferably, described two visual point images by rgb format carry out resolution ratio normalization,Comprise: two visual point images to rgb format carry out convergent-divergent, formation resolution ratio is 1280*360Two visual point images.
Preferably, in the function L (x, y) corresponding with described left visual point image, the span of xBe 0～639 pixel, the span of y is 0～359 pixel; With described right visual point image pairIn the function R (x, y) answering, the span of x is 640～1279 pixels, the span of yIt is 0～359 pixel.
Preferably, described left visual point image and described right visual point image carried out to same formatCut apart, comprising: described left visual point image and described right visual point image are cut apart respectively,Obtain subimage, the width of described subimage is 160 pixels, is highly 40 pixels.
The embodiment of the present invention also provides the depth map acquisition device of two visual point images, comprising:Normalization module, for carrying out resolution ratio normalizing by two visual point images of RGB rgb formatChange; Format converting module, for changing described two visual point images after resolution ratio normalizationFor yuv format, and extract acquisition brightness signal Y and divide spirogram, wherein said Y divides spirogramComprise left visual point image and right visual point image; Cut apart module, for by described left visual point imageAnd described right visual point image carries out cutting apart of same format, with in described left visual point image and instituteState the multiple subimages pair that form the mutual correspondence in position in right visual point image; Motion search module,For for each subimage pair, using the subimage of described left visual point image as for working asThe current subimage of front operation; Make center and the described current son of default search patternThe pixel that meets preset rules of image overlaps; Taking the subimage of described right visual point image asSearch subimage; In the same position of described search subimage, described search pattern is set; ?In described search subimage, move described search pattern along the width of described search subimageCenter, all calculate described search when the center of mobile described search pattern at every turnSubimage be positioned at described search pattern scope pixel absolute error and, when described absolutelyStop the motion search of described search pattern to error with while being less than predetermined threshold value, and record instituteState the distance that move search pattern center; All subimages to all complete motion search itAfter, utilize the distance of record to obtain left depth map; For each subimage pair, with describedThe subimage of left visual point image is as search subimage, according to the side of obtaining of described left depth mapFormula obtains right depth map; Depth map acquisition module, for respectively by described left depth map and instituteThe depth value of stating in right depth map is mapped in 0～255 scope, obtains the dark of two visual point imagesDegree figure;
Wherein, described absolute error and the following method of employing obtain:
Set up xoy coordinate system, in described xoy coordinate system, with described left visual point image pairThe function of answering is designated as L (x, y), and the function corresponding with described right visual point image is designated as R (x, y);
When each center of moving described search pattern, calculating described search subimage is positioned atThe absolute error of the pixel within the scope of described search pattern and computing formula be,
f(k)＝Σ(x∈1-9)Σ(y∈1-5)|M(x,y)*L(x,y)-M(x+k,y)*R(x,y)|；
Wherein M(x,y)For width is 9 pixels, be highly the search pattern of 5 pixels, f (k) isDifference mean value, the distance that move the center that k is described search pattern;
Wherein, the described pixel that meets preset rules comprises: the pixel that pixel value is non-vanishingPoint or pixel coordinate value meet the pixel of x%2==0&&y%2==0.
Preferably, described device adopts on-site programmable gate array FPGA device to realize.
Depth map acquisition methods and the device of two visual point images that the embodiment of the present invention provides,Divide left visual point image and right visual point image in spirogram to cut apart the brightness signal Y of extractionFor multiple subimages, and subimage and right visual point image meta in left visual point imageThe subimage of putting mutual correspondence forms subimage pair, in the degree of depth of obtaining two visual point imagesIn figure process,, walk abreast to multiple subimages to entering to as hunting zone using subimageRow motion search, forms left depth map and right depth map, and utilizes the depth map obtainingCarry out the rear depth map that forms two visual point images of depth value mapping. The embodiment of the present inventionDepth map acquisition methods and the device of two visual point images, obtain two with of the prior artThe method of the depth map of visual point image is compared, and data processing amount is little, and data processing is real-timeProperty high, be easy to realize, more can meet user and obtain the reality of the depth map of two visual point imagesBorder demand.
Brief description of the drawings
Fig. 1 shows the stream of the depth map acquisition methods of two visual point images in the embodiment of the present inventionCheng Tu;
Fig. 2 shows the structural representation of search pattern in the embodiment of the present invention;
Fig. 3 shows the mark effect schematic diagram of pixel in the invention process;
Fig. 4 shows the knot of the depth map acquisition device of two visual point images in the embodiment of the present inventionStructure schematic diagram.
Detailed description of the invention
Also by reference to the accompanying drawings the present invention is done further in detail below by specific embodimentDescribe.
A kind of depth map acquisition methods of two visual point images is provided in embodiments of the invention,As shown in Figure 1, main treatment step comprises:
Step S11: two visual point images of RGB rgb format are carried out to resolution ratio normalization;
Step S12: two visual point images after resolution ratio normalization are converted to yuv format,And extract acquisition brightness signal Y and divide spirogram, wherein Y divides spirogram to comprise left visual point image and the right sideVisual point image;
Step S13: left visual point image and right visual point image are carried out to cutting apart of same format,To form position multiple subimages of correspondence mutually in left visual point image and right visual point imageRight;
Step S14: parallel to multiple subimages to carrying out motion search, obtain left depth mapAnd right depth map;
Step S15: respectively the depth value in left depth map and right depth map is mapped to 0～255In scope, obtain the depth map of two visual point images.
In the embodiment of the present invention, the brightness signal Y of extraction is divided to the left viewpoint figure in spirogramPicture and right visual point image are divided into multiple subimages, and subgraph in left visual point imageThe subimage that picture is mutual corresponding with position in right visual point image forms subimage pair, is obtainingGet in the depth map process of two visual point images, using subimage to as hunting zone, andRow to carrying out motion search, forms left depth map and right depth map to multiple subimages,And utilize the depth map obtaining to carry out the rear degree of depth that forms two visual point images of depth value mappingFigure. Depth map acquisition methods and the device of two visual point images of the embodiment of the present invention, withThe method of the depth map that obtains two visual point images of the prior art is compared, data processingMeasure littlely, data processing real-time is high, is easy to realize, and more can meet user and obtain two and lookThe actual demand of the depth map of dot image.
In the embodiment of the present invention, two visual point images of rgb format are carried out to resolution ratio normalization,Be convenient to the post-processed of two visual point images, concrete normalized resolution ratio can be according to actual needAsk setting, preferably right in two visual point images of rgb format are offered an explanation to normalizationTwo visual point images of rgb format carry out convergent-divergent, and to form resolution ratio be 1280*360 two looksDot image.
Two visual point images that the resolution ratio wherein forming is 1280*360 are that finger widths is 1280Pixel is highly two visual point images of 360 pixels.
In frequency normalization, two visual point images are zoomed to 1280*360 can expire simultaneouslyThe demand of foot image definition and image processing speed.
Two visual point images of the rgb format of wish processing are carried out after resolution ratio normalization, willTwo visual point images of rgb format are converted to yuv format.
Wherein, a kind of colour coding method that YUV is adopted by eurovision system. ?In modern vitascan, conventionally adopt three pipe colour cameras or colored CCD video cameraCarry out capture, then the colour picture signal of obtaining is obtained after color separation, difference amplification correctionTo RGB, then obtain brightness signal Y and two colour difference signal R through matrixer-Y (being U), B-Y (being V), last transmitting terminal divides brightness and three signals of aberrationDo not encode, send with same channel. The method for expressing of this color is exactly so-calledYUV color space represent. The importance that adopts YUV color space is its brightness letterNumber Y separates with carrier chrominance signal U, V.
In the embodiment of the present invention, two visual point images after resolution ratio normalization are converted to YUVAfter form, the Y of extraction divides two visual point images after resolution ratio and the resolution ratio normalization of spirogramResolution ratio identical.
For example, after resolution ratio normalization, the resolution ratio of two visual point images is 1280*360,It is still 1280*360 that the Y that yuv format extracts after changing divides the resolution ratio of spirogram.
From be converted to two visual point images of yuv format, extract and obtain Y and divide the concrete of spirogramStep comprises: Y divides spirogram to be made up of Y component value, and Y component value is by two of rgb formatIn visual point image, conversion obtains, and wherein the transfer algorithm of Y component value is:Y=R*I+G*J+B*K; Wherein, Y represents Y component value, and R, G and B represent respectivelyThe red component R value of a pixel, green component G value in two visual point images of GRB formAnd blue component B value, I, J and K represent respectively red component R value, green component G valueAnd the weighted value of blue component B value, and meet: I+J+K=1 and I:J:K ≈ 1:4.3:2.7.
Obtaining after left visual point image and right visual point image, adopting same format respectively a left side to be lookedDot image and right visual point image are cut apart, and on left visual point image and right visual point image, divideDo not form multiple subimages, and the left visual point image subgraph identical with right visual point image positionPicture forms subimage pair.
In the embodiment of the present invention, walk abreast described multiple subimages are obtained to a left side to carrying out motion searchDepth map comprises: for each subimage pair, do with the subimage of described left visual point imageFor the current subimage for current operation; Make center and the institute of default search patternThe pixel that meets preset rules of stating current subimage overlaps; With described right visual point imageSubimage is search subimage; In the same position of described search subimage, described search is setTemplate; Described in moving along the width of described search subimage in described search subimageAll calculate when the center of at every turn moving described search pattern the center of search patternDescribed search subimage is positioned at the difference mean value of the pixel of described search pattern scope,In the time that being less than predetermined threshold value, described difference mean value stops the motion search of described search pattern,And record the distance that move described search pattern center; All subimages are to all completing fortuneAfter moving search, utilize the distance of record to obtain left depth map; Parallel to described multiple subgraphsPicture, to carrying out motion search, obtains right depth map and comprises: for each subimage pair, withThe subimage of described right visual point image is as search subimage, according to obtaining of described left depth mapThe mode of getting obtains right depth map.
In the embodiment of the present invention, further illustrate how by left visual point image and the right side are lookedThe motion search of dot image obtains left depth map and right depth map.
Specifically be exemplified as: that sets two visual point images to rgb format carries out frequency normalizingThe frequency of two visual point images after change is 1280*360, for looking in the left visual point image and the right side that obtainDot image is set up xoy coordinate system, the function corresponding with left visual point image in xoy coordinate systemBe designated as L (x, y), the function corresponding with right visual point image is designated as R (x, y), with left visual point imageIn corresponding function L (x, y), the span of x is 0～639 pixel, and the span of y is0～359 pixel; In the function R (x, y) corresponding with right visual point image, the span of x is640～1279 pixels, the span of y is 0～359 pixel.
In the time that left visual point image and right visual point image are carried out to cutting apart of same format, particularlyLeft visual point image and right visual point image can be cut apart respectively, be obtained subimage, subgraphThe width of picture is 160 pixels, is highly 40 pixels.
In left visual point image, form 36 number of sub images, in right visual point image, form 36Number of sub images, the subimage at the same position place of left visual point image and right visual point image forms sonImage pair, for example, in left visual point image, x coordinate is (0～159) pixel, ordinate is (0～39)X coordinate in the subimage of pixel and right visual point image is (640～799) pixel, ordinateFor the subimage of (0～39) pixel forms subimage pair.
The computational methods of difference mean value in the embodiment of the present invention, comprising: set up xoy coordinateSystem, in described xoy coordinate system, the function corresponding with left visual point image is designated as L (x, y),The function corresponding with right visual point image is designated as R (x, y); The centre bit of each mobile search templateWhile putting, calculating search subimage is positioned at the difference mean value of the pixel of search pattern scopeComputing formula is:
f(k)＝Σ(x∈1-9)Σ(y∈1-5)|M(x,y)*L(x,y)-M(x+k,y)*R(x,y)|; WhereinM(xy,For width is 9 pixels, be highly the search pattern of 5 pixels, f (k) is that difference is averageValue, the distance that move the center that k is search pattern.
Preserve in k value to buffer register buffer, the buffer wherein adopting is also160x40, the value of concrete k is set according to the size of subimage, for example, can be 50 of +/-Pixel, or be 45 pixels of +/-.
The pixel that meets further preset rules comprises: the pixel that pixel value is non-vanishingOr pixel coordinate value meets the pixel of x%2==0&&y%2==0.
As shown in Figure 2, be the non-vanishing pixel of pixel value when meeting the pixel of preset rulesWhen some in,, dividing at Y the pixel value that meets this regular pixel in spirogram is 1, utilizes 9x5Search pattern (being that width is 9 pixels, is highly the search pattern of 5 pixels), searchThe pixel that in the coverage of template, pixel value is 1 participates in calculating, the ground that pixel value is 0Side does not participate in calculating. Utilize the template of 9x5, with the current subimage in left visual point imageEach pixel value is that 1 pixel overlaps with the central point of search pattern, at right visual point imageThe same position of subimage (be search subimage) of same position on place phase appositionThe search pattern of formula, and utilize search pattern to move along the width of search subimage, whenThe difference mean value of the pixel in the search subimage that search pattern covers is less than predetermined threshold valueOr the difference mean value that calculates of judgement hour, stops search, and be recorded in right viewpointThe distance that in image, move the center of search pattern. Utilize identical method to left viewpointAll pixel values in image are not that 0 pixel correspondence is moved in right visual point imageSearch, and the distance that move the center of record searching template respectively. Utilize left viewpoint figureThe distance of the record that the non-vanishing pixel of all pixel values in picture is corresponding forms the left degree of depthFigure.
Adopt identical method, taking right visual point image as main, in left visual point image, move and searchRope, the distance of record searching template center position record, and utilize the distance of record to generate rightDepth map.
Particularly, each range data is recorded in the spatial cache identical with subimage formIn.
For higher raising search efficiency, after verification experimental verification, find in practical application notNeed to all carry out search arithmetic to whole subimage, in subimage, have some pixels passableNeed not search for, just passable by the direct interpolation of value of data before and after getting, meet default rulePixel can also meet for pixel coordinate value the pixel of x%2==0&&y%2==0Point, as only searched for the pixel that is labeled as Y in Fig. 3, other are labeled as the point of xDo not search for, do not do the point of searching for, supplement and out just can by front and back point data average interpolation.
In the embodiment of the present invention, respectively the depth value in left depth map and right depth map is mapped to0～255 is specifically as follows, and the depth value of setting in the depth map obtaining is-50～+ 50,-50～0 correspondence mappings to 0～127,1～50 is mapped to 128～255.
In the embodiment of the present invention, left visual point image and right visual point image are cut apart respectively to left viewpointSubimage in image and the multiple subimages pair of right visual point image neutron image construction, carrying outWhen motion search, parallel multiple subimages are carried out to motion search, in left visual point imageAll subimages are all confirmed as current subimage; While determining left depth map, all current sonsImage coordinate with its respectively corresponding search subimage walk abreast and carry out motion search.
In the process of depth map of obtaining two visual point images, can utilize FPGA to realize rightThe parallel motion search that multiple subimages are right, improves the data processing speed of obtaining depth map.
A kind of depth map acquisition device of two visual point images is also provided in the embodiment of the present invention,Comprise as shown in Figure 4:
Normalization module 21, for dividing two visual point images of RGB rgb formatDistinguish rate normalization;
Format converting module 22, for being converted to two visual point images after resolution ratio normalizationYuv format, and extract acquisition brightness signal Y and divide spirogram, wherein Y divides spirogram to comprise a left sideVisual point image and right visual point image;
Cut apart module 23, for left visual point image and right visual point image being carried out to same formatCut apart, to form position multiple sons of correspondence mutually in left visual point image and right visual point imageImage pair;
Motion search module 24, searches moving described multiple subimages for parallelRope, obtains left depth map and right depth map;
Depth map acquisition module 25, for respectively by the degree of depth of left depth map and right depth mapValue is mapped in 0～255 scope, obtains the depth map of two visual point images.
Preferably, the depth map acquisition device of the embodiment of the present invention adopts field-programmable gate arrayRow FPGA device is realized, and utilizes FPGA to realize the right parallel motion to multiple subimages and searchesRope, improves the data processing speed of obtaining depth map.
The foregoing is only the preferred embodiments of the present invention, be not limited to the present invention,For a person skilled in the art, the present invention can have various modifications and variations. AllWithin the spirit and principles in the present invention, any amendment of doing, be equal to replacement, improvement etc.,Within all should being included in protection scope of the present invention.
Claims (7)
1. the depth map acquisition methods of liang visual point image, is characterized in that, comprising:
Two visual point images of RGB rgb format are carried out to resolution ratio normalization;
Described two visual point images after resolution ratio normalization are converted to yuv format, and carryGet acquisition brightness signal Y and divide spirogram, wherein said Y divides spirogram to comprise left visual point image and the right sideVisual point image;
Described left visual point image and described right visual point image are carried out to cutting apart of same format, withIn described left visual point image and described right visual point image, form position multiple sons of correspondence mutuallyImage pair;
For each subimage pair, using the subimage of described left visual point image as for working asThe current subimage of front operation; Make center and the described current son of default search patternThe pixel that meets preset rules of image overlaps; Taking the subimage of described right visual point image asSearch subimage; In the same position of described search subimage, described search pattern is set; ?In described search subimage, move described search pattern along the width of described search subimageCenter, all calculate described search when the center of mobile described search pattern at every turnSubimage be positioned at described search pattern scope pixel absolute error and, when described absolutelyStop the motion search of described search pattern to error with while being less than predetermined threshold value, and record instituteState the distance that move search pattern center; All subimages to all complete motion search itAfter, utilize the distance of record to obtain left depth map;
For each subimage pair, sub as search using the subimage of described left visual point imageImage, obtains right depth map according to the obtain manner of described left depth map;
Respectively the depth value in described left depth map and described right depth map is mapped to 0～255In scope, obtain the depth map of two visual point images;
Wherein, described absolute error and the following method of employing obtain:
Set up xoy coordinate system, in described xoy coordinate system, with described left visual point image pairThe function of answering is designated as L (x, y), and the function corresponding with described right visual point image is designated as R (x, y);
When each center of moving described search pattern, calculating described search subimage is positioned atThe absolute error of the pixel within the scope of described search pattern and computing formula be,
f(k)＝∑(x∈1-9)∑(y∈1-5)|M(x,y)*L(x,y)-M(x+k,y)*R(x,y)|；
Wherein M(x,y)For width is 9 pixels, be highly the search pattern of 5 pixels, f (k) isDifference mean value, the distance that move the center that k is described search pattern;
Wherein, the described pixel that meets preset rules comprises: the pixel that pixel value is non-vanishingPoint or pixel coordinate value meet the pixel of x%2=0&&y%2=0.
2. method according to claim 1, is characterized in that, described extraction obtainsY divides spirogram, comprising:
Described Y divides spirogram to be made up of Y component value, and described Y component value is by rgb formatIn two visual point images, conversion obtains, and wherein the transfer algorithm of Y component value is:
Y＝R*I+G*J+B*K；
Wherein, Y represents Y component value, and R, G and B represent that respectively two of GRB form looksThe red component R value of a pixel, green component G value and blue component B value in dot image,I, J and K represent respectively red component R value, green component G value and blue component B valueWeighted value, and meet: I+J+K=1 and I:J:K ≈ 1:4.3:2.7.
3. method according to claim 1, is characterized in that, described by RGBTwo visual point images of form carry out resolution ratio normalization, comprising:
Two visual point images to rgb format carry out convergent-divergent, and formation resolution ratio is 1280*360Two visual point images.
4. method according to claim 3, is characterized in that, with described left viewpointIn function L (x, y) corresponding to image, the span of x is 0～639 pixel, the value model of yEnclosing is 0～359 pixel;
In the function R (x, y) corresponding with described right visual point image, the span of x is640～1279 pixels, the span of y is 0～359 pixel.
5. method according to claim 4, is characterized in that, by described left viewpointImage and described right visual point image carry out cutting apart of same format, comprising:
Described left visual point image and described right visual point image are cut apart respectively, obtained subgraphPicture, the width of described subimage is 160 pixels, is highly 40 pixels.
6. the depth map acquisition device of liang visual point image, is characterized in that, comprising:
Normalization module, for differentiating two visual point images of RGB rgb formatRate normalization;
Format converting module, for changing described two visual point images after resolution ratio normalizationFor yuv format, and extract acquisition brightness signal Y and divide spirogram, wherein said Y divides spirogramComprise left visual point image and right visual point image;
Cut apart module, for described left visual point image and described right visual point image are carried out identicalCutting apart of form, to form position phase in described left visual point image and described right visual point imageCorresponding multiple subimages pair mutually;
Motion search module, for for each subimage pair, with described left visual point imageSubimage as the current subimage for current operation; Make default search pattern inHeart position overlaps with the pixel that meets preset rules of described current subimage; With the described right sideThe subimage of visual point image is search subimage; Identical bits at described search subimage installsPut described search pattern; In described search subimage along the width side of described search subimageTo the center of mobile described search pattern, the centre bit of mobile described search pattern at every turnWhile putting, all calculate described search subimage be positioned at described search pattern scope pixel absolutelyTo error and, stop described search pattern when described absolute error with while being less than predetermined threshold valueMotion search, and record the distance that move described search pattern center; All subimagesAfter all completing motion search, utilize the distance of record to obtain left depth map; For eachNumber of sub images pair, using the subimage of described left visual point image as search subimage, according to instituteThe obtain manner of stating left depth map obtains right depth map;
Depth map acquisition module, for respectively by described left depth map and described right depth mapDepth value be mapped in 0～255 scope, obtain the depth map of two visual point images;
Wherein, described absolute error and the following method of employing obtain:
Set up xoy coordinate system, in described xoy coordinate system, with described left visual point image pairThe function of answering is designated as L (x, y), and the function corresponding with described right visual point image is designated as R (x, y);
When each center of moving described search pattern, calculating described search subimage is positioned atThe absolute error of the pixel within the scope of described search pattern and computing formula be,
f(k)＝∑(x∈1-9)∑(y∈1-5)|M(x,y)*L(x,y)-M(x+k,y)*R(x,y)|；
Wherein M(x,y)For width is 9 pixels, be highly the search pattern of 5 pixels, f (k) isDifference mean value, the distance that move the center that k is described search pattern;
Wherein, the described pixel that meets preset rules comprises: the pixel that pixel value is non-vanishingPoint or pixel coordinate value meet the pixel of x%2=0&&y%2=0.
7. device according to claim 6, is characterized in that, described device adoptsOn-site programmable gate array FPGA device is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410064382.4A CN103780897B (en) | 2014-02-25 | 2014-02-25 | Depth map acquisition methods and the device of two visual point images |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410064382.4A CN103780897B (en) | 2014-02-25 | 2014-02-25 | Depth map acquisition methods and the device of two visual point images |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103780897A CN103780897A (en) | 2014-05-07 |
| CN103780897B true CN103780897B (en) | 2016-05-11 |
Family
ID=50572670
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410064382.4A Expired - Fee Related CN103780897B (en) | 2014-02-25 | 2014-02-25 | Depth map acquisition methods and the device of two visual point images |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103780897B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105704473A (en) * | 2016-02-01 | 2016-06-22 | 李应樵 | Method for quickly generating images for multi-view three-dimensional display screen by use of templates |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101662695A (en) * | 2009-09-24 | 2010-03-03 | 清华大学 | Method and device for acquiring virtual viewport |
| CN101720047A (en) * | 2009-11-03 | 2010-06-02 | 上海大学 | Method for acquiring range image by stereo matching of multi-aperture photographing based on color segmentation |
| CN102026012A (en) * | 2010-11-26 | 2011-04-20 | 清华大学 | Generation method and device of depth map through three-dimensional conversion to planar video |
| CN102254348A (en) * | 2011-07-25 | 2011-11-23 | 北京航空航天大学 | Block matching parallax estimation-based middle view synthesizing method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8848038B2 (en) * | 2010-07-09 | 2014-09-30 | Lg Electronics Inc. | Method and device for converting 3D images |
-
2014
- 2014-02-25 CN CN201410064382.4A patent/CN103780897B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101662695A (en) * | 2009-09-24 | 2010-03-03 | 清华大学 | Method and device for acquiring virtual viewport |
| CN101720047A (en) * | 2009-11-03 | 2010-06-02 | 上海大学 | Method for acquiring range image by stereo matching of multi-aperture photographing based on color segmentation |
| CN102026012A (en) * | 2010-11-26 | 2011-04-20 | 清华大学 | Generation method and device of depth map through three-dimensional conversion to planar video |
| CN102254348A (en) * | 2011-07-25 | 2011-11-23 | 北京航空航天大学 | Block matching parallax estimation-based middle view synthesizing method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103780897A (en) | 2014-05-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104301677B (en) | The method and device monitored towards the panoramic video of large scene | |
| Guttmann et al. | Semi-automatic stereo extraction from video footage | |
| CN101271578B (en) | Depth sequence generation method of technology for converting plane video into stereo video | |
| CN105814875B (en) | Selecting camera pairs for stereo imaging | |
| CN102857739B (en) | Distributed overall view monitoring system and method thereof | |
| CN101742349B (en) | Method for expressing three-dimensional scenes and television system thereof | |
| CN103248906B (en) | Method and system for acquiring depth map of binocular stereo video sequence | |
| CN103606151B (en) | Based on the virtual geographical scene method for auto constructing on a large scale of imaging point cloud | |
| CN102223553A (en) | Method for converting two-dimensional video into three-dimensional video automatically | |
| CN103337094A (en) | Method for realizing three-dimensional reconstruction of movement by using binocular camera | |
| CN103345736A (en) | Virtual viewpoint rendering method | |
| CN101662695B (en) | Method and device for acquiring virtual viewport | |
| CN102223545B (en) | Rapid multi-view video color correction method | |
| CN103024420B (en) | 2D-3D (two-dimension to three-dimension) conversion method for single images in RGBD (red, green and blue plus depth) data depth migration | |
| Fan et al. | Vivid-DIBR based 2D–3D image conversion system for 3D display | |
| CN104301706A (en) | Synthetic method for improving naked eye stereoscopic display effect | |
| CN103871037A (en) | Method and apparatus for color transfer between images | |
| CN109345444A (en) | The super-resolution stereo-picture construction method of depth perception enhancing | |
| CN103780897B (en) | Depth map acquisition methods and the device of two visual point images | |
| US20140347352A1 (en) | Apparatuses, methods, and systems for 2-dimensional and 3-dimensional rendering and display of plenoptic images | |
| KR101281003B1 (en) | Image processing system and method using multi view image | |
| CN102708570A (en) | Method and device for obtaining depth map | |
| GB2585197A (en) | Method and system for obtaining depth data | |
| CN102938845B (en) | Real-time virtual viewpoint generation method based on perspective projection | |
| CN106028018B (en) | Real scene shooting double vision point 3D method for optimizing video and system towards naked eye 3D display |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160511 Termination date: 20180225 |