CN108986155A - The depth estimation method and estimation of Depth equipment of multi-view image - Google Patents
The depth estimation method and estimation of Depth equipment of multi-view image Download PDFInfo
- Publication number
- CN108986155A CN108986155A CN201710413057.8A CN201710413057A CN108986155A CN 108986155 A CN108986155 A CN 108986155A CN 201710413057 A CN201710413057 A CN 201710413057A CN 108986155 A CN108986155 A CN 108986155A
- Authority
- CN
- China
- Prior art keywords
- pixel
- depth
- parameter
- image
- condition
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
Abstract
The invention discloses a kind of depth estimation methods of multi-view image and estimation of Depth equipment.This method comprises: to Same Scene multiple images each pixel distribution characterization depth and characterize Surface by Tangent Plane Method to parameter;Each image is executed as present image: a) cost function value being selected to meet the pixel of condition as sub-pixel;B) parameter of the surrounding pixel for meeting condition around sub-pixel is updated;C) in the preset range of the parameter of surrounding pixel, random search obtains local optimum;D) surrounding pixel that cost function value meets condition is increased as sub-pixel;E) repeat b), c), d) step, until not meeting the surrounding pixel of condition;F) parameter is generated at random for each pixel in present image, in the case where meeting condition, the parameter current of the pixel is replaced with the parameter generated at random;G) step a)-f is repeated), until meeting condition;H) depth value that the depth parameter of pixel each in present image determines is determined as depth value.
Description
Technical field
This invention relates generally to technical field of image processing.It can be with lower specifically, the present invention relates to one kind
Computation complexity estimates that the multi-view image of any scene includes the method and apparatus of the depth of high-definition picture.
Background technique
In recent years, with the fast development of camera technique, can obtain largely has high-resolution image.From these
The recovery that scene or object dimensional structure are carried out in image, has great importance for many computer applications.For example, joy
Pleasure, augmented reality, paleovolcanic structure, robot etc..
The estimation of Depth of multi-view image is the committed step of the three-dimensional modeling based on image, has been widely studied.More views
The depth estimation method of point image can be generally divided into two classes: voxel-based method and the method based on cloud.Based on voxel
Method good effect can be obtained on the three-dimensional modeling of single body, but be difficult to be extended in the scene of large scale.
Method based on cloud can be applied in arbitrary scene, but due to usually requiring optimization graph model, thus have higher
Computation complexity.
Therefore, the present invention is directed to any scene especially high-definition picture can be estimated with lower computation complexity
Depth.
Summary of the invention
It has been given below about brief overview of the invention, in order to provide about the basic of certain aspects of the invention
Understand.It should be appreciated that this summary is not an exhaustive overview of the invention.It is not intended to determine pass of the invention
Key or pith, nor is it intended to limit the scope of the present invention.Its purpose only provides certain concepts in simplified form,
Taking this as a prelude to a more detailed description discussed later.
The purpose of the present invention is to propose to one kind, and any scene especially high-resolution can be estimated with lower computation complexity
The method and apparatus of the depth of rate image.
To achieve the goals above, according to an aspect of the invention, there is provided a kind of estimation of Depth of multi-view image
Method, this method comprises: each pixel in each image into the multiple images of Same Scene is distributed for characterizing the picture
One depth parameter of plain depth and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions;It will be in described multiple images
Each image execute following processing as present image: a) cost function value that imaging consistency is measured in selection meets first
The pixel of part is as sub-pixel;B) depth parameter and method of the surrounding pixel for meeting second condition around sub-pixel are updated
To parameter;C) in the predetermined value range locating for the depth parameter being updated and normal dimensions of surrounding pixel, made a reservation for
The random search of number, to obtain the depth parameter of surrounding pixel and the local optimum of normal dimensions;D) by cost function value
It is sub-pixel that the surrounding pixel for meeting first condition, which increases,;E) repeat it is above-mentioned b), c), d) step, until without meeting first
The surrounding pixel of condition;F) depth parameter and two normal dimensions are generated at random for each pixel in present image, and
And in the case where meeting third condition, by the current depth parameter of the pixel and the normal dimensions depth parameter generated at random
It is replaced with normal dimensions;G) repeat the above steps a)-f), until meeting fourth condition;H) by pixel each in present image
The depth value that depth parameter determines is determined as the depth value of the pixel.
According to another aspect of the present invention, a kind of estimation of Depth equipment of multi-view image is provided, which includes:
Central processing unit CPU is configured as controlling: each pixel distribution in each image into the multiple images of Same Scene
For characterize the pixel depth a depth parameter and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions;By institute
The each image stated in multiple images executes following processing as present image: a) cost function of imaging consistency is measured in selection
Value meets the pixel of first condition as sub-pixel;B) surrounding pixel for meeting second condition around sub-pixel is updated
Depth parameter and normal dimensions;C) the predetermined value range locating for the depth parameter being updated and normal dimensions of surrounding pixel
It is interior, the random search of pre-determined number is carried out, to obtain the depth parameter of surrounding pixel and the local optimum of normal dimensions;D) will
It is sub-pixel that the surrounding pixel that cost function value meets first condition, which increases,;E) repeat it is above-mentioned b), c), d) step, until not having
There is the surrounding pixel for meeting first condition;F) depth parameter and two methods are generated at random for each pixel in present image
To parameter, and in the case where meeting third condition, the current depth parameter and normal dimensions of the pixel are generated with random
Depth parameter and normal dimensions replace;G) repeat the above steps a)-f), until meeting fourth condition;It h) will be in present image
The depth value that the depth parameter of each pixel determines is determined as the depth value of the pixel.
In addition, according to another aspect of the present invention, additionally providing a kind of storage medium.The storage medium includes that machine can
The program code of reading, when executing said program code on information processing equipment, said program code makes at the information
Equipment is managed to execute according to the above method of the present invention.
In addition, in accordance with a further aspect of the present invention, additionally providing a kind of program product.Described program product includes that machine can
The instruction of execution, when executing described instruction on information processing equipment, described instruction executes the information processing equipment
According to the above method of the present invention.
Detailed description of the invention
Referring to reference to the accompanying drawing to the explanation of the embodiment of the present invention, the invention will be more easily understood it is above and
Other objects, features and advantages.Component in attached drawing is intended merely to show the principle of the present invention.In the accompanying drawings, identical or class
As technical characteristic or component will be indicated using same or similar appended drawing reference.In attached drawing:
Fig. 1 shows the flow chart of the depth estimation method of the multi-view image of embodiment according to the present invention;
Fig. 2 shows the flow charts that depth value determines processing;
Fig. 3 shows the structural block diagram of the estimation of Depth equipment of the multi-view image of embodiment according to the present invention;With
And
Fig. 4 shows the schematic frame for the computer that can be used for implementing the method and apparatus of embodiment according to the present invention
Figure.
Specific embodiment
Exemplary embodiment of the invention is described in detail hereinafter in connection with attached drawing.It rises for clarity and conciseness
See, does not describe all features of actual implementation mode in the description.It should be understood, however, that developing any this reality
Much decisions specific to embodiment must be made during embodiment, to realize the objectives of developer,
For example, meeting restrictive condition those of related to system and business, and these restrictive conditions may be with embodiment
It is different and change.In addition, it will also be appreciated that although development is likely to be extremely complex and time-consuming, to benefit
For those skilled in the art of present disclosure, this development is only routine task.
Here, and also it should be noted is that, in order to avoid having obscured the present invention because of unnecessary details, in the accompanying drawings
Illustrate only with closely related apparatus structure and/or processing step according to the solution of the present invention, and be omitted and the present invention
The little other details of relationship.In addition, it may also be noted that being described in an attached drawing of the invention or a kind of embodiment
Elements and features can be combined with elements and features shown in one or more other attached drawings or embodiment.
The basic idea of the invention is that using the scene for adapting to any scale based on the method for cloud, while using at random
It generates seed point and carries out avoiding the need for optimization graph model based on the mode that the selectivity of random search extends based on seed point
And introduce higher computation complexity.In addition, the present invention is when calculating cost function value using the throwing of multiple pixels in window
Shadow replaces the projection of the single center pixel of window to solve the problems, such as the perspective distortion of window.In addition, invention introduces four kinds
The method of Corrected Depth value, to further increase the accuracy of depth value.
The process of the depth estimation method of the multi-view image of embodiment according to the present invention is described below with reference to Fig. 1.
Fig. 1 shows the flow chart of the depth estimation method of the multi-view image of embodiment according to the present invention.Such as Fig. 1 institute
Show, this method comprises the following steps: each pixel distribution in each image into the multiple images of Same Scene is for table
Levy the pixel depth a depth parameter and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions (step S1);It will
Each image in described multiple images executes depth value as present image and determines processing (step S2).
In step sl, each pixel in each image into the multiple images of Same Scene is distributed for characterizing this
One depth parameter of pixel depth and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions.
As noted previously, as the present invention is based on the modes of cloud, it is possible to adapt to the scene of any scale.For same
Scene takes pictures to the scene by one group of camera with different location, orientation, obtains multiple images, such multiple figures
Picture referred to as multi-view image.The above-mentioned one group camera with different location, orientation can be translated by a camera, rotate to difference
Position, orientation are to substitute.Application scenarios of the invention are that multi-view image has already passed through calibration, i.e., the internal reference of known camera and outside
Ginseng.The internal reference of camera refers to that the parameter of camera itself, such as focal length, the outer ginseng of camera are related to the direction of camera, such as rotation, translation
Deng.Internal reference, outer ginseng, the depth of pixel based on camera, the three-dimensional point in three-dimensional space can be determined according to pixel, can also be incited somebody to action
Three-dimensional point projects in image, can will be on the pixel projection to another image on an image.
Current invention assumes that each pixel in image has a supporting plane (tangent plane), the parameter of the supporting plane by
The depth and normal direction of the pixel determine.The cost function value that imaging consistency is calculated by the way of supporting plane, utilizes imaging
Consistency constraint obtains the depth value of pixel.
The hypothesis of imaging consistency are as follows: the feature of subpoint of the three-dimensional point on different perspectives image in scene
(such as color, gray scale) is consistent as far as possible.Specifically, the current pixel of an image (i.e. present image) is (i.e. right
Its calculate imaging consistency cost function value pixel) cost function value be in the predetermined window centered on current pixel
Multiple first pixels are corresponding, multiple second pixels in each adjacent image of present image and the multiple first picture
The average value of feature difference between element.
There are multiple first pixels in predetermined window (such as rectangle) centered on current pixel in present image, currently
There are multiple second pixels corresponding with above-mentioned multiple first pixels in each adjacent image of image.It should be noted that multiple second pictures
Element not necessarily constitutes a window (such as rectangle) similar with predetermined window.Calculate multiple second pixels in adjacent image
The average value of feature difference between multiple first pixels in present image, and it is directed to the feature difference of each adjacent image
Average value is averaged, the cost function value of the current pixel as present image.
As described above, multi-view image is calibrated, it is known that the internal reference of camera, outer ginseng.Therefore, the multiple second pixel
Can according to and the internal reference of the corresponding camera of present image, outer ginseng, the internal reference of camera corresponding with adjacent image, outer ginseng, currently
Depth parameter, the normal dimensions of pixel determine.It is assumed that multiple first pixels in predetermined window are because of image local similitude
And supporting plane parameter having the same, therefore, using the depth parameter of current pixel and normal dimensions as multiple first pictures
Element shared depth parameter and normal dimensions.
In the conventional technology, it when determining corresponding multiple second pixels according to the current pixel of present image, uses
Mode be the depth parameter based on current pixel, current pixel is mapped to the subpoint in adjacent image, selects the projection
The pixel in predetermined window centered on point, as multiple second pixels.Multiple second pixels acquired in this way are actually not
The necessarily corresponding subpoint of multiple first pixels, thus lead to the perspective distortion problem of window.It, will using method of the invention
Multiple first pixels project to obtain the second pixel respectively, can solve the perspective distortion problem of window.
The depth parameter of pixel and the initial value of normal dimensions are obtained by distribution.Specifically, to multiple figures of Same Scene
Each pixel in each image as in distributes a depth parameter and two normal dimensions.The depth of pixel only needs a depth
Degree parameter can characterize, and can acquire mutually between the depth and depth parameter of pixel.The Surface by Tangent Plane Method of pixel is to needs two
Normal dimensions can just characterize, and the Surface by Tangent Plane Method of pixel can acquire mutually between two normal dimensions.
Depth parameter, normal dimensions all have maximum value and minimum value.It therefore, in one embodiment, can be in minimum
In section defined by value and maximum value, it is randomly assigned depth parameter and normal dimensions.
In another embodiment, can use inferred motion structure technology is that several pixels distribute depth parameter.This is
Because inferred motion structure technology is capable of determining that the three-dimensional coordinate of sparse pixel, and then can determine that out depth, obtain in this way
Depth is more accurate.It in addition, being randomly assigned normal direction parameter for above-mentioned several pixels, and is the pixel in addition to above-mentioned several pixels
It is randomly assigned depth parameter and normal dimensions.
The purpose of step S1 is to obtain initial value.These initial values in addition to inferred motion structure technology obtain it is reliable other than,
The inevitable inaccuracy mostly of the initial value that obtains at random, but necessarily have the initial value of small part (an at least one) parameter relatively subject to
Really.Reliably and accurate initial value based on this part, based on, start depth value in step s 2 and determine processing.
Fig. 2 shows the flow charts that depth value determines processing.
As shown in Fig. 2, depth value according to the present invention determines processing using each image in multiple images as current figure
As executing following steps:
A) cost function value for measuring imaging consistency is selected to meet the pixel of first condition as sub-pixel;
B) depth parameter and normal dimensions of the surrounding pixel for meeting second condition around sub-pixel are updated;
C) in the predetermined value range locating for the depth parameter being updated and normal dimensions of surrounding pixel, made a reservation for
The random search of number, to obtain the depth parameter of surrounding pixel and the local optimum of normal dimensions;
D) surrounding pixel that cost function value meets first condition is increased as sub-pixel;
E) repeat it is above-mentioned b), c), d) step, until not meeting the surrounding pixel of first condition;
F) depth parameter and two normal dimensions are generated at random for each pixel in present image, and meeting
In the case where third condition, the current depth parameter of the pixel and normal dimensions are joined with the depth parameter and normal direction generated at random
Number replaces;
G) repeat the above steps a)-f), until meeting fourth condition;
H) depth value that the depth parameter of pixel each in present image determines is determined as to the depth value of the pixel.
Due to having been obtained for the depth parameter of pixel and the initial value of normal dimensions in step sl, it is possible to according to such as
The upper cost function value calculating method, calculates the current depth parameter and the corresponding cost function of normal dimensions of each pixel
Value.In step a), selection meets first condition as measured the cost function value of imaging consistency and be less than predetermined cost threshold value
Pixel is as sub-pixel.Predetermined cost threshold value can be specified by those skilled in the art, as with reliable parameter value
The standard of reliable pixel.
Next, carrying out selective extension from sub-pixel, increasing the quantity of reliable pixel.
If the surrounding pixel around sub-pixel meets second condition, if surrounding pixel is in the depth with sub-pixel
Corresponding cost function value is less than surrounding pixel at current depth parameter and when normal dimensions pairs when parameter and normal dimensions
The cost function value answered then illustrates that surrounding pixel is suitable for that (immanent cause is for the depth parameter with sub-pixel and normal dimensions
The locally coherence of image), the depth parameter of surrounding pixel and normal dimensions are updated to the depth parameter of sub-pixel respectively
And normal dimensions, realize the selectivity extension of the parameter of sub-pixel.If above-mentioned second condition is unsatisfactory for, retain surrounding picture
The current depth parameter and normal dimensions of element.
It should be noted here that: surrounding pixel can be the surface of sub-pixel, underface, front-left, four of front-right
Pixel (four neighborhoods) is also possible to eight pixels (eight neighborhood) enclosed around sub-pixel one, is located at image side in surrounding pixel
In the case where near edge, surrounding pixel can be the partial pixel in aforementioned four pixel or eight pixels.
In addition, surrounding pixel is not limited to non-seed pixel, surrounding pixel itself is also possible to sub-pixel.As long as surrounding
The current parameter value of pixel meets second condition, i.e., when having the parameter value of the sub-pixel of (four neighborhoods or eight neighborhood) nearby
Cost function value is smaller, then its parameter can be updated to the parameter value of sub-pixel.
The parameter value of the sub-pixel updated by surrounding pixel is not necessarily the optimal value of surrounding pixel,
In order to further increase the accuracy of estimation of Depth, random search is carried out, in step c) to seek more preferably parameter value.It should infuse
Meaning: step c) targeted object is its parameter value by those of the parameter value update of sub-pixel surrounding pixel.
For the depth parameter being updated and normal dimensions of surrounding pixel, respectively centered on parameter, allow in parameter
Constant interval in carry out randomly rather than search for (generate random point) with having tendentiousness, and calculate the corresponding cost of random point
Functional value.The minimum parameter value of cost function value in parameter current and random point is retained.
Random search can repeat pre-determined number.Moreover, in a preferred embodiment, each random search can be upper
Range is searched further on the basis of random search.For example, the predetermined value range of random search is that last time is random every time
The half of the value range of search, centered on the depth parameter being updated and normal dimensions of surrounding pixel.For example, all
The current depth parameter for enclosing pixel is 5, and the depth parameter of the sub-pixel near surrounding pixel is 7.It is computed, surrounding pixel
Cost function value when depth parameter is 7 is less than the cost function value when depth parameter of surrounding pixel is 5, therefore, surrounding picture
The depth parameter of element is updated to 7.Centered on 7, for example, random search is carried out in [3,11] section (siding-to-siding block length 8),
Obtain random point 6.5.It is computed, the cost function value when depth parameter of surrounding pixel is 6.5 is less than the depth of surrounding pixel
Cost function value when parameter is 7.Therefore, random search is taken turns by this, the depth parameter of surrounding pixel is updated to 6.5.Under
When random search, centered on 6.5, random search is carried out in [4.5,8.5] section (siding-to-siding block length 4), obtain with
Machine point 6.6.It is computed, the depth parameter that the cost function value when depth parameter of surrounding pixel is 6.6 is less than surrounding pixel is
Cost function value when 6.5.Therefore, random search is taken turns by this, the depth parameter of surrounding pixel is updated to 6.6.Assuming that predetermined
Number is 2, then the local optimum of the depth parameter of surrounding pixel is 6.6.
Judge whether the corresponding cost function value of current parameter value of surrounding pixel meets first condition in step d), i.e.,
The Rule of judgment of sub-pixel, as cost function value is less than predetermined cost threshold value.If it is determined that meet first condition, then it will be all
Enclosing pixel to increase is sub-pixel.In this way, by step b), c), d) by the parameter spread of sub-pixel to portion picture
Element, and the surrounding pixel (pixel that surrounding pixel may include sub-pixel type) of sub-pixel is on the basis of expanding value
Local optimum is obtained by random search, the surrounding pixel for having updated parameter value increases the number of sub-pixel.
In step e), repeat it is above-mentioned b), c), d) step, until not meeting the surrounding pixel of first condition.That is,
It is extended again based on all sub-pixels for increasing novel species sub-pixel, random search, obtains local optimum, increases kind
The process of sub-pixel.During this period, sub-pixel is continuously increased, and the parameter value of sub-pixel and surrounding surrounding pixel is continuous
It obtains updating optimization.
Due to above-mentioned circulation be all from reliable sub-pixel, the parameter updated improves depth
Spend the accuracy of estimation.However, since most of or whole initial parameter values are from randomization, so the distribution of seed may
It is very average, it is also possible to concentrate very much.For example, there are the foreground objects such as big tree, stone in current scene, and big tree and stone that
This is separate.If initial sub-pixel concentrates on setting greatly, then the surrounding pixel updated according to sub-pixel is also on big tree
Or near big tree, without regard to stone.Stone and its parameter value of neighbouring pixel, which are just not much reliable initial value, not to be had yet
Adequately optimized.
In order to avoid such situation, increase the range of sub-pixel distribution, is every in present image in step f)
A pixel generates a depth parameter and two normal dimensions at random.This randomization may introduce stone and its neighbouring kind
Sub-pixel.In the case where meeting third condition, by the current depth parameter of the pixel and the normal dimensions depth generated at random
It spends parameter and normal dimensions replaces.The third condition includes: that the pixel has the depth parameter being randomly assigned and normal direction ginseng
Corresponding cost function value is less than the pixel with current depth parameter and whens normal dimensions corresponding cost function value when number.
In step g), repeat the above steps a)-f), until meeting fourth condition.
As step f) is repeated, the distribution of sub-pixel is more and more extensive, and the quantity of sub-pixel is more and more.With
Step a)-f) repetition, from sub-pixel update surrounding pixel it is more and more, the parameter value updated is more and more accurate.
Fourth condition includes: that iteration reaches sub-pixel and last time in pre-determined number or current iteration in step a)
The number difference of sub-pixel in iteration in step a) is less than quantity threshold.Quantity threshold can be referred to by those skilled in the art
It is fixed.
By above step a)-f), enough sub-pixel and updated surrounding pixel, the ginseng of pixel can be obtained
Numerical value is more accurate.
In step h), the depth value that the depth parameter of pixel each in present image determines can be determined as the pixel
Depth value.
In order to remove erroneous estimation and noise, the accuracy of estimation of Depth is further increased, before step h), may be used also
To carry out adjusting for depth operation.
The present invention provides four kinds of adjusting for depth operations, can select at least one therein.
The first amendment operation is the depth parameter for being unsatisfactory for first condition for removing each pixel in multiple images.?
That is removal cost function value is less than the depth parameter of predetermined cost threshold value.Predetermined cost threshold value can be by those skilled in the art
Member is specified, can be equal to cost threshold value when judging whether it is sub-pixel, can also slightly below judge whether it is sub-pixel
When cost threshold value.
Second of amendment operation is to project the depth parameter of present image and normal dimensions by the internal reference of camera, outer ginseng
Whether the subpoint into adjacent image reduces according to cost function value, to determine whether with the depth parameter of present image and
Normal dimensions update the depth and normal direction of subpoint in adjacent image.
Specifically, execute following processing for each image in described multiple images as present image: based on it is current
The internal reference of the corresponding camera of image, outer ginseng, the internal reference of camera corresponding with the adjacent image of present image, outer ginseng, present image
In each pixel depth parameter, by subpoint of each pixel projection into adjacent image in present image;Meeting
In the case where fifth condition, the depth parameter of subpoint is updated with the depth parameter and normal dimensions of the pixel in present image
And normal dimensions.
Wherein, fifth condition includes: subpoint in depth parameter and normal dimensions with the pixel in present image
When corresponding cost function value be less than subpoint corresponding cost in the current depth parameter and normal dimensions with subpoint
Functional value.
Further, it is also possible to, using adjacent image as present image, execute step a)-c herein on basis).In this way, phase
By the extension of sub-pixel, the random search of surrounding pixel in adjacent image, the more accurate depth of adjacent image can be obtained
Value.
The theoretical basis of the third amendment operation is depth of the same three-dimensional point in different perspectives image in scene
It should be consistent.Therefore, insecure estimation of Depth value is removed using the method that depth consistency is examined.
Specifically, execute following processing for each image in described multiple images as present image: based on it is current
The internal reference of the corresponding camera of image, it is outer ginseng, camera corresponding with multiple adjacent images of present image internal reference, it is outer ginseng, currently
The depth parameter of each pixel in image, by each pixel projection in present image to the subpoint in multiple adjacent images
And obtain the first depth of subpoint;The second of the current depth parameter decision of the first depth and subpoint of compared projections point is deep
Depth difference between degree;If the depth difference of the subpoint of the corresponding predetermined number of the pixel is greater than predetermined difference threshold value,
Then remove the depth parameter of the pixel.Otherwise it is assumed that having passed through the inspection of depth consistency.
Wherein, the depth difference of the first depth and the second depth can be the absolute of the first depth and the difference of the second depth
Value is also possible to the absolute value of the difference of the first depth and the second depth divided by the first depth or the second depth.
4th kind of amendment operation is filtered to the depth map being made of identified depth value, to remove noise and fill out
Fill cavity.Filter is, for example, median filter, two-sided filter etc..
The estimation of Depth equipment of the multi-view image of embodiment according to the present invention is described next, with reference to Fig. 3.
Fig. 3 shows the structural block diagram of the estimation of Depth equipment of the multi-view image of embodiment according to the present invention.Such as
Shown in Fig. 3, the estimation of Depth equipment 300 of multi-view image according to the present invention includes: central processing unit CPU, is configured as
Control: each pixel in each image into the multiple images of Same Scene distributes one for characterizing the pixel depth
Depth parameter and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions;Each image in described multiple images is made
Execute following processing for present image: a) select the cost function value of measurement imaging consistency meet the pixel of first condition as
Sub-pixel;B) depth parameter and normal dimensions of the surrounding pixel for meeting second condition around sub-pixel are updated;C) exist
In predetermined value range locating for the depth parameter being updated and normal dimensions of surrounding pixel, searching at random for pre-determined number is carried out
Rope, to obtain the depth parameter of surrounding pixel and the local optimum of normal dimensions;D) cost function value is met into first condition
Surrounding pixel increase be sub-pixel;E) repeat it is above-mentioned b), c), d) step, until not meeting picture around first condition
Element;F) depth parameter and two normal dimensions are generated at random for each pixel in present image, and meeting third
In the case where condition, by the current depth parameter of the pixel and the normal dimensions depth parameter and normal dimensions generation generated at random
It replaces;G) repeat the above steps a)-f), until meeting fourth condition;H) depth parameter of pixel each in present image is determined
Depth value be determined as the depth value of the pixel.
In one embodiment, the cost function value of the current pixel of present image be with it is pre- centered on current pixel
Determine corresponding multiple first pixels in window, multiple second pixels in each adjacent image of present image and described more
The average value of feature difference between a first pixel, wherein the multiple second pixel is according to camera corresponding with present image
Internal reference, outer ginseng, the internal reference of camera corresponding with adjacent image, outer ginseng, the depth parameter of current pixel, normal dimensions determine.
In one embodiment, the CPU is further configured to control: being randomly assigned the depth parameter and normal direction ginseng
Number.
In one embodiment, the CPU is further configured to control: being several pictures using inferred motion structure technology
Element distribution depth parameter, and normal direction parameter is randomly assigned for several pixels;For the pixel in addition to several pixels
It is randomly assigned the depth parameter and normal dimensions.
In one embodiment, the first condition includes that cost function value is less than predetermined cost threshold value.
In one embodiment, the second condition includes: surrounding pixel in depth parameter and method with sub-pixel
Corresponding cost function value is less than surrounding pixel with current depth parameter and whens normal dimensions corresponding cost when to parameter
Functional value.
In one embodiment, the CPU is further configured to control: the depth parameter of surrounding pixel and normal direction are joined
Number is updated to the depth parameter and normal dimensions of sub-pixel respectively.
In one embodiment, the predetermined value range of each random search is the one of the value range of last time random search
Half, centered on the depth parameter being updated and normal dimensions of surrounding pixel.
In one embodiment, the third condition includes: that the pixel has the depth parameter being randomly assigned and normal direction
Corresponding cost function value is less than the pixel with current depth parameter and whens normal dimensions corresponding cost function when parameter
Value.
In one embodiment, the fourth condition includes: that iteration reaches in pre-determined number or current iteration in step a)
Sub-pixel and last iteration in the number difference of sub-pixel in step a) be less than quantity threshold.
In one embodiment, the CPU is further configured to control: each pixel in removal multiple images is not
Meet the depth parameter of first condition.
In one embodiment, the CPU is further configured to control: each image in described multiple images is made
Following processing is executed for present image: the neighbor map of internal reference, outer ginseng and present image based on camera corresponding with present image
As the depth parameter of each pixel in the internal reference of corresponding camera, outer ginseng, present image, by each pixel in present image
Project to the subpoint in adjacent image;In the case where meeting fifth condition, joined with the depth of the pixel in present image
Several and normal dimensions update the depth parameter and normal dimensions of subpoint.
In one embodiment, the fifth condition includes: subpoint in the depth with the pixel in present image
Corresponding cost function value is less than subpoint in current depth parameter and normal direction ginseng with subpoint when parameter and normal dimensions
Corresponding cost function value when number.
In one embodiment, the CPU is further configured to control: using adjacent image as present image, executing
Step a)-c).
In one embodiment, the CPU is further configured to control: each image in described multiple images is made
Following processing is executed for present image: multiple phases of internal reference, outer ginseng and present image based on camera corresponding with present image
The depth parameter of the internal reference of the corresponding camera of adjacent image, outer ginseng, each pixel in present image, by each of present image
Pixel projection is to the subpoint in multiple adjacent images and obtains the first depth of subpoint;First depth of compared projections point with
The depth difference between the second depth that the current depth parameter of subpoint determines;If the throwing of the corresponding predetermined number of the pixel
The depth difference of shadow point is greater than predetermined difference threshold value, then removes the depth parameter of the pixel.
In one embodiment, the CPU is further configured to control: to the depth being made of identified depth value
Figure is filtered, to remove noise and filling cavity.
Due to included processing in the estimation of Depth equipment 300 of multi-view image according to the present invention be described above
Multi-view image depth estimation method in processing in included each step it is similar, therefore for simplicity,
This omits the detailed description of these processing.
In addition, it is still necessary to, it is noted that each component devices, unit can be by softwares, firmware, hard in above equipment here
The mode of part or combinations thereof is configured.It configures workable specific means or mode is well known to those skilled in the art,
This is repeated no more.In the case where being realized by software or firmware, from storage medium or network to specialized hardware structure
Computer (such as general purpose computer 400 shown in Fig. 4) installation constitutes the program of the software, which is being equipped with various journeys
When sequence, it is able to carry out various functions etc..
Fig. 4 shows the schematic frame for the computer that can be used for implementing the method and apparatus of embodiment according to the present invention
Figure.
In Fig. 4, central processing unit (CPU) 401 is according to the program stored in read-only memory (ROM) 402 or from depositing
The program that storage part 408 is loaded into random access memory (RAM) 403 executes various processing.In RAM 403, also according to need
Store the data required when CPU 401 executes various processing etc..CPU 401, ROM 402 and RAM 403 are via bus
404 are connected to each other.Input/output interface 405 is also connected to bus 404.
Components described below is connected to input/output interface 405: importation 406 (including keyboard, mouse etc.), output section
Divide 407 (including display, such as cathode-ray tube (CRT), liquid crystal display (LCD) etc. and loudspeakers etc.), storage section
408 (including hard disks etc.), communications portion 409 (including network interface card such as LAN card, modem etc.).Communications portion 409
Communication process is executed via network such as internet.As needed, driver 410 can be connected to input/output interface 405.
Detachable media 411 such as disk, CD, magneto-optic disk, semiconductor memory etc., which can according to need, is installed in driver
On 410, so that the computer program read out is mounted to as needed in storage section 408.
It is such as removable from network such as internet or storage medium in the case where series of processes above-mentioned by software realization
Unload the program that the installation of medium 411 constitutes software.
It will be understood by those of skill in the art that this storage medium be not limited to it is shown in Fig. 4 be wherein stored with program,
Separately distribute with equipment to provide a user the detachable media 411 of program.The example of detachable media 411 includes disk
(including floppy disk (registered trademark)), CD (including compact disc read-only memory (CD-ROM) and digital versatile disc (DVD)), magneto-optic disk
(including mini-disk (MD) (registered trademark)) and semiconductor memory.Alternatively, storage medium can be ROM 402, storage section
Hard disk for including in 408 etc., wherein computer program stored, and user is distributed to together with the equipment comprising them.
The present invention also proposes a kind of program product of instruction code for being stored with machine-readable.Described instruction code is by machine
When device reads and executes, method that above-mentioned embodiment according to the present invention can be performed.
Correspondingly, it is also wrapped for carrying the storage medium of the program product of the above-mentioned instruction code for being stored with machine-readable
It includes in disclosure of the invention.The storage medium includes but is not limited to floppy disk, CD, magneto-optic disk, storage card, memory stick etc.
Deng.
In the description above to the specific embodiment of the invention, for the feature a kind of embodiment description and/or shown
It can be used in one or more other embodiments in a manner of same or similar, with the feature in other embodiment
It is combined, or the feature in substitution other embodiment.
It should be emphasized that term "comprises/comprising" refers to the presence of feature, element, step or component when using herein, but simultaneously
It is not excluded for the presence or additional of one or more other features, element, step or component.
In addition, method of the invention be not limited to specifications described in time sequencing execute, can also according to it
His time sequencing, concurrently or independently execute.Therefore, the execution sequence of method described in this specification is not to this hair
Bright technical scope is construed as limiting.
Although being had been disclosed above by the description to specific embodiments of the present invention to the present invention, it answers
The understanding, above-mentioned all embodiments and example are exemplary, and not restrictive.Those skilled in the art can be in institute
Design is to various modifications of the invention, improvement or equivalent in attached spirit and scope of the claims.These modification, improve or
Person's equivalent should also be as being to be considered as included in protection scope of the present invention.
Note
1. a kind of depth estimation method of multi-view image, comprising:
Each pixel in each image into the multiple images of Same Scene is distributed for characterizing the pixel depth
One depth parameter and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions;
Following processing is executed using each image in described multiple images as present image:
A) cost function value for measuring imaging consistency is selected to meet the pixel of first condition as sub-pixel;
B) depth parameter and normal dimensions of the surrounding pixel for meeting second condition around sub-pixel are updated;
C) in the predetermined value range locating for the depth parameter being updated and normal dimensions of surrounding pixel, made a reservation for
The random search of number, to obtain the depth parameter of surrounding pixel and the local optimum of normal dimensions;
D) surrounding pixel that cost function value meets first condition is increased as sub-pixel;
E) repeat it is above-mentioned b), c), d) step, until not meeting the surrounding pixel of first condition;
F) depth parameter and two normal dimensions are generated at random for each pixel in present image, and meeting
In the case where third condition, the current depth parameter of the pixel and normal dimensions are joined with the depth parameter and normal direction generated at random
Number replaces;
G) repeat the above steps a)-f), until meeting fourth condition;
H) depth value that the depth parameter of pixel each in present image determines is determined as to the depth value of the pixel.
2. note 1 as described in method, wherein the cost function value of the current pixel of present image be with current pixel
Centered on predetermined window in multiple first pixels are corresponding, multiple second pictures in each adjacent image of present image
The average value of the plain feature difference between the multiple first pixel, wherein the multiple second pixel according to present image
The internal reference of corresponding camera, outer ginseng, the internal reference of camera corresponding with adjacent image, outer ginseng, the depth parameter of current pixel, normal direction
Parameter determines.
3. the method as described in note 1, wherein the allocation step includes: to be randomly assigned the depth parameter and normal direction
Parameter.
4. the method as described in note 1, wherein it is several that the allocation step, which includes: using inferred motion structure technology,
Pixel distributes depth parameter, and is randomly assigned normal direction parameter for several pixels;For the picture in addition to several pixels
Element is randomly assigned the depth parameter and normal dimensions.
5. the method as described in note 1, wherein the first condition includes that cost function value is less than predetermined cost threshold value.
6. the method as described in note 1, wherein the second condition includes: surrounding pixel in the depth with sub-pixel
It spends parameter and whens normal dimensions corresponding cost function value is less than surrounding pixel with current depth parameter and when normal dimensions
Corresponding cost function value.
7. the method as described in note 6, wherein the update step includes: by the depth parameter and normal direction of surrounding pixel
Parameter is updated to the depth parameter and normal dimensions of sub-pixel respectively.
8. the method as described in note 1, wherein the predetermined value range of each random search is taking for last time random search
It is worth the half of range, centered on the depth parameter being updated and normal dimensions of surrounding pixel.
9. the method as described in note 1, wherein the third condition includes: that the pixel has the depth being randomly assigned
Corresponding cost function value is less than the pixel corresponding with current depth parameter and when normal dimensions when parameter and normal dimensions
Cost function value.
10. the method as described in note 1, wherein the fourth condition includes: that iteration reaches pre-determined number or this changes
The number difference of sub-pixel in sub-pixel and last iteration in generation in step a) in step a) is less than quantity threshold.
11. the method as described in note 1, further includes: remove each pixel in multiple images is unsatisfactory for first condition
Depth parameter.
12. the method as described in note 1, further includes:
Following processing is executed using each image in described multiple images as present image:
Based on and the internal reference of the corresponding camera of present image, outer ginseng, camera corresponding with the adjacent image of present image
The depth parameter of internal reference, outer ginseng, each pixel in present image, by each pixel projection in present image to adjacent image
In subpoint;
In the case where meeting fifth condition, is updated and thrown with the depth parameter and normal dimensions of the pixel in present image
The depth parameter and normal dimensions of shadow point.
13. the method as described in note 12, wherein the fifth condition includes: subpoint in present image
Corresponding cost function value is less than subpoint in the current depth with subpoint when the depth parameter and normal dimensions of the pixel
Corresponding cost function value when parameter and normal dimensions.
14. the method as described in note 12, further includes: using adjacent image as present image, execute step a)-c).
15. the method as described in note 1, further includes:
Following processing is executed using each image in described multiple images as present image:
Based on and the corresponding camera of present image internal reference, outer ginseng, phase corresponding with multiple adjacent images of present image
The depth parameter of the internal reference of machine, outer ginseng, each pixel in present image, by each pixel projection in present image to multiple
Subpoint in adjacent image simultaneously obtains the first depth of subpoint;
The depth difference between the second depth that first depth of compared projections point and the current depth parameter of subpoint determine
It is different;
If the depth difference of the subpoint of the corresponding predetermined number of the pixel is greater than predetermined difference threshold value, the picture is removed
The depth parameter of element.
16. the method as described in note 1, further includes: the depth map being made of identified depth value is filtered, with
Remove noise and filling cavity.
17. a kind of estimation of Depth equipment of multi-view image, comprising:
Central processing unit CPU is configured as controlling:
Each pixel in each image into the multiple images of Same Scene is distributed for characterizing the pixel depth
One depth parameter and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions;
Following processing is executed using each image in described multiple images as present image:
A) cost function value for measuring imaging consistency is selected to meet the pixel of first condition as sub-pixel;
B) depth parameter and normal dimensions of the surrounding pixel for meeting second condition around sub-pixel are updated;
C) in the predetermined value range locating for the depth parameter being updated and normal dimensions of surrounding pixel, made a reservation for
The random search of number, to obtain the depth parameter of surrounding pixel and the local optimum of normal dimensions;
D) surrounding pixel that cost function value meets first condition is increased as sub-pixel;
E) repeat it is above-mentioned b), c), d) step, until not meeting the surrounding pixel of first condition;
F) depth parameter and two normal dimensions are generated at random for each pixel in present image, and meeting
In the case where third condition, the current depth parameter of the pixel and normal dimensions are joined with the depth parameter and normal direction generated at random
Number replaces;
G) repeat the above steps a)-f), until meeting fourth condition;
H) depth value that the depth parameter of pixel each in present image determines is determined as to the depth value of the pixel.
18. note 1 as described in equipment, wherein the cost function value of the current pixel of present image be with current picture
Multiple first pixels in predetermined window centered on element are corresponding, multiple second in each adjacent image of present image
The average value of feature difference between pixel and the multiple first pixel, wherein the multiple second pixel is schemed according to current
As the internal reference of corresponding camera, outer ginseng, the internal reference of camera corresponding with adjacent image, outer ginseng, the depth parameter of current pixel, method
It is determined to parameter.
19. the equipment as described in note 1, wherein the CPU is further configured to control:
Following processing is executed using each image in described multiple images as present image:
Based on and the internal reference of the corresponding camera of present image, outer ginseng, camera corresponding with the adjacent image of present image
The depth parameter of internal reference, outer ginseng, each pixel in present image, by each pixel projection in present image to adjacent image
In subpoint;
In the case where meeting fifth condition, is updated and thrown with the depth parameter and normal dimensions of the pixel in present image
The depth parameter and normal dimensions of shadow point.
20. the equipment as described in note 1, wherein the CPU is further configured to control:
Following processing is executed using each image in described multiple images as present image:
Based on and the corresponding camera of present image internal reference, outer ginseng, phase corresponding with multiple adjacent images of present image
The depth parameter of the internal reference of machine, outer ginseng, each pixel in present image, by each pixel projection in present image to multiple
Subpoint in adjacent image simultaneously obtains the first depth of subpoint;
The depth difference between the second depth that first depth of compared projections point and the current depth parameter of subpoint determine
It is different;
If the depth difference of the subpoint of the corresponding predetermined number of the pixel is greater than predetermined difference threshold value, the picture is removed
The depth parameter of element.
Claims (10)
1. a kind of depth estimation method of multi-view image, comprising:
Each pixel in each image into the multiple images of Same Scene distributes one for characterizing the pixel depth
Depth parameter and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions;
Following processing is executed using each image in described multiple images as present image:
A) cost function value for measuring imaging consistency is selected to meet the pixel of first condition as sub-pixel;
B) depth parameter and normal dimensions of the surrounding pixel for meeting second condition around sub-pixel are updated;
C) in the predetermined value range locating for the depth parameter being updated and normal dimensions of surrounding pixel, pre-determined number is carried out
Random search, to obtain the depth parameter of surrounding pixel and the local optimum of normal dimensions;
D) surrounding pixel that cost function value meets first condition is increased as sub-pixel;
E) repeat it is above-mentioned b), c), d) step, until not meeting the surrounding pixel of first condition;
F) depth parameter and two normal dimensions are generated at random for each pixel in present image, and meeting third
In the case where condition, by the current depth parameter of the pixel and the normal dimensions depth parameter and normal dimensions generation generated at random
It replaces;
G) repeat the above steps a)-f), until meeting fourth condition;
H) depth value that the depth parameter of pixel each in present image determines is determined as to the depth value of the pixel.
2. the method for claim 1, wherein the cost function value of the current pixel of present image be with current pixel
Centered on predetermined window in multiple first pixels are corresponding, multiple second pictures in each adjacent image of present image
The average value of the plain feature difference between the multiple first pixel, wherein the multiple second pixel according to present image
The internal reference of corresponding camera, outer ginseng, the internal reference of camera corresponding with adjacent image, outer ginseng, the depth parameter of current pixel, normal direction
Parameter determines.
3. the method for claim 1, wherein the second condition includes: surrounding pixel in the depth with sub-pixel
It spends parameter and whens normal dimensions corresponding cost function value is less than surrounding pixel with current depth parameter and when normal dimensions
Corresponding cost function value.
4. the method for claim 1, wherein the third condition includes: that the pixel has the depth being randomly assigned
Corresponding cost function value is less than the pixel corresponding with current depth parameter and when normal dimensions when parameter and normal dimensions
Cost function value.
5. the method for claim 1, wherein the fourth condition includes: that iteration reaches pre-determined number or current iteration
The number difference of sub-pixel in sub-pixel and last iteration in middle step a) in step a) is less than quantity threshold.
6. the method as described in claim 1, further includes: each pixel in removal multiple images is unsatisfactory for first condition
Depth parameter.
7. the method as described in claim 1, further includes:
Following processing is executed using each image in described multiple images as present image:
Based on and the internal reference of the corresponding camera of present image, outer ginseng, camera corresponding with the adjacent image of present image internal reference,
The depth parameter of outer ginseng, each pixel in present image, by each pixel projection in present image into adjacent image
Subpoint;
In the case where meeting fifth condition, subpoint is updated with the depth parameter and normal dimensions of the pixel in present image
Depth parameter and normal dimensions.
8. the method as described in claim 1, further includes:
Following processing is executed using each image in described multiple images as present image:
Based on and the internal reference of the corresponding camera of present image, outer ginseng, camera corresponding with multiple adjacent images of present image
The depth parameter of internal reference, outer ginseng, each pixel in present image, by each pixel projection in present image to multiple adjacent
Subpoint in image simultaneously obtains the first depth of subpoint;
The depth difference between the second depth that first depth of compared projections point and the current depth parameter of subpoint determine;
If the depth difference of the subpoint of the corresponding predetermined number of the pixel is greater than predetermined difference threshold value, the pixel is removed
Depth parameter.
9. the method as described in claim 1, further includes:
The depth map being made of identified depth value is filtered, to remove noise and filling cavity.
10. a kind of estimation of Depth equipment of multi-view image, comprising:
Central processing unit CPU is configured as controlling:
Each pixel in each image into the multiple images of Same Scene distributes one for characterizing the pixel depth
Depth parameter and for characterize the pixel Surface by Tangent Plane Method to two normal dimensions;
Following processing is executed using each image in described multiple images as present image:
A) cost function value for measuring imaging consistency is selected to meet the pixel of first condition as sub-pixel;
B) depth parameter and normal dimensions of the surrounding pixel for meeting second condition around sub-pixel are updated;
C) in the predetermined value range locating for the depth parameter being updated and normal dimensions of surrounding pixel, pre-determined number is carried out
Random search, to obtain the depth parameter of surrounding pixel and the local optimum of normal dimensions;
D) surrounding pixel that cost function value meets first condition is increased as sub-pixel;
E) repeat it is above-mentioned b), c), d) step, until not meeting the surrounding pixel of first condition;
F) depth parameter and two normal dimensions are generated at random for each pixel in present image, and meeting third
In the case where condition, by the current depth parameter of the pixel and the normal dimensions depth parameter and normal dimensions generation generated at random
It replaces;
G) repeat the above steps a)-f), until meeting fourth condition;
H) depth value that the depth parameter of pixel each in present image determines is determined as to the depth value of the pixel.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710413057.8A CN108986155B (en) | 2017-06-05 | 2017-06-05 | Depth estimation method and depth estimation apparatus for multi-viewpoint image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710413057.8A CN108986155B (en) | 2017-06-05 | 2017-06-05 | Depth estimation method and depth estimation apparatus for multi-viewpoint image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108986155A true CN108986155A (en) | 2018-12-11 |
CN108986155B CN108986155B (en) | 2021-12-07 |
Family
ID=64502534
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710413057.8A Active CN108986155B (en) | 2017-06-05 | 2017-06-05 | Depth estimation method and depth estimation apparatus for multi-viewpoint image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108986155B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111104641A (en) * | 2019-12-10 | 2020-05-05 | 重庆大学 | Method for identifying crystal grains by computer in three-dimensional space |
CN111815696A (en) * | 2019-04-11 | 2020-10-23 | 曜科智能科技(上海)有限公司 | Depth map optimization method, device, equipment and medium based on semantic instance segmentation |
CN113435465A (en) * | 2020-03-20 | 2021-09-24 | 阿里巴巴集团控股有限公司 | Image processing and intelligent control method and equipment |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102609977A (en) * | 2012-01-12 | 2012-07-25 | 浙江大学 | Depth integration and curved-surface evolution based multi-viewpoint three-dimensional reconstruction method |
CN102903096A (en) * | 2012-07-04 | 2013-01-30 | 北京航空航天大学 | Monocular video based object depth extraction method |
US20140192158A1 (en) * | 2013-01-04 | 2014-07-10 | Microsoft Corporation | Stereo Image Matching |
US9177381B2 (en) * | 2010-12-22 | 2015-11-03 | Nani Holdings IP, LLC | Depth estimate determination, systems and methods |
CN106548466A (en) * | 2015-09-16 | 2017-03-29 | 富士通株式会社 | The method and apparatus of three-dimensional reconstruction object |
-
2017
- 2017-06-05 CN CN201710413057.8A patent/CN108986155B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9177381B2 (en) * | 2010-12-22 | 2015-11-03 | Nani Holdings IP, LLC | Depth estimate determination, systems and methods |
CN102609977A (en) * | 2012-01-12 | 2012-07-25 | 浙江大学 | Depth integration and curved-surface evolution based multi-viewpoint three-dimensional reconstruction method |
CN102903096A (en) * | 2012-07-04 | 2013-01-30 | 北京航空航天大学 | Monocular video based object depth extraction method |
US20140192158A1 (en) * | 2013-01-04 | 2014-07-10 | Microsoft Corporation | Stereo Image Matching |
CN106548466A (en) * | 2015-09-16 | 2017-03-29 | 富士通株式会社 | The method and apparatus of three-dimensional reconstruction object |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111815696A (en) * | 2019-04-11 | 2020-10-23 | 曜科智能科技(上海)有限公司 | Depth map optimization method, device, equipment and medium based on semantic instance segmentation |
CN111815696B (en) * | 2019-04-11 | 2023-08-22 | 曜科智能科技(上海)有限公司 | Depth map optimization method, device, equipment and medium based on semantic instance segmentation |
CN111104641A (en) * | 2019-12-10 | 2020-05-05 | 重庆大学 | Method for identifying crystal grains by computer in three-dimensional space |
CN113435465A (en) * | 2020-03-20 | 2021-09-24 | 阿里巴巴集团控股有限公司 | Image processing and intelligent control method and equipment |
Also Published As
Publication number | Publication date |
---|---|
CN108986155B (en) | 2021-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7990379B2 (en) | System and method for coronary segmentation and visualization | |
US11734879B2 (en) | Graphics processing using directional representations of lighting at probe positions within a scene | |
CN105164728B (en) | For mixing the apparatus and method in real border | |
CN110874864A (en) | Method, device, electronic equipment and system for obtaining three-dimensional model of object | |
TWI536318B (en) | Depth measurement quality enhancement | |
EP3367334B1 (en) | Depth estimation method and depth estimation apparatus of multi-view images | |
US10275930B2 (en) | Combined intensity projection | |
US11049270B2 (en) | Method and apparatus for calculating depth map based on reliability | |
CN107980150A (en) | Three dimensions is modeled | |
JP3557982B2 (en) | Optical flow estimation method | |
CN113140011A (en) | Infrared thermal imaging monocular vision distance measurement method and related assembly | |
CN108986155A (en) | The depth estimation method and estimation of Depth equipment of multi-view image | |
CN110544294B (en) | Dense three-dimensional reconstruction method based on panoramic video | |
JP7285834B2 (en) | Three-dimensional reconstruction method and three-dimensional reconstruction apparatus | |
WO2006064478A1 (en) | High quality accurate surface triangulation from a simplex mesh | |
CN107292826A (en) | Image processing equipment, image processing method and image processing apparatus | |
GB2563596A (en) | System and method for modeling a three dimensional space based on a two dimensional image | |
CN109377552B (en) | Image occlusion calculating method, device, calculating equipment and storage medium | |
US10402970B2 (en) | Model-based segmentation of an anatomical structure | |
US20150145861A1 (en) | Method and arrangement for model generation | |
CN112233149A (en) | Scene flow determination method and device, storage medium and electronic device | |
US20220156887A1 (en) | Kernel reshaping-powered splatting-based efficient image space lens blur | |
CN110378948B (en) | 3D model reconstruction method and device and electronic equipment | |
CN113284081B (en) | Depth map super-resolution optimization method and device, processing equipment and storage medium | |
US11253154B2 (en) | Method and system for magnetic resonance imaging |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |