CN109146782A - A kind of full-view image joining method and system - Google Patents
A kind of full-view image joining method and system Download PDFInfo
- Publication number
- CN109146782A CN109146782A CN201810811294.4A CN201810811294A CN109146782A CN 109146782 A CN109146782 A CN 109146782A CN 201810811294 A CN201810811294 A CN 201810811294A CN 109146782 A CN109146782 A CN 109146782A
- Authority
- CN
- China
- Prior art keywords
- moving target
- image
- spliced
- target
- full
- 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.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000005304 joining Methods 0.000 title claims abstract description 19
- 238000001514 detection method Methods 0.000 claims abstract description 19
- 230000011218 segmentation Effects 0.000 claims abstract description 14
- 238000005520 cutting process Methods 0.000 claims abstract description 5
- 238000012937 correction Methods 0.000 claims description 10
- 230000000386 athletic effect Effects 0.000 claims description 8
- 238000013135 deep learning Methods 0.000 claims description 4
- 230000008034 disappearance Effects 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 16
- 238000005265 energy consumption Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Studio Devices (AREA)
Abstract
The present invention relates to a kind of full-view image joining method and systems, this method includes by adjustment of image to be spliced in the same coordinate system, acquisition moving target is split to image to be spliced, motion state judgement and conspicuousness detection are carried out to obtained moving target, tax power is carried out to moving target according to judgement and testing result, cutting energy-optimised algorithm using figure combines the weight assigned to carry out splicing line detection, obtains optimal splicing line.The present invention divides network by example can be more complete to moving Object Segmentation, cuts energy-optimised algorithm using figure, in conjunction with the weight of moving target, splicing line can be made around moving target, generate the full-view image of clean background, dislocation-free.
Description
Technical field
The present invention relates to image procossing and technical field of computer vision more particularly to a kind of full-view image joining method and
System.
Background technique
Full-view image is widely used in outer scene, because its angular field of view is big such as streetscape map, scenic spot indoors
Roaming figure and navigation etc..Wherein, full-view image splicing is key technology and hot research problem.In static scene, to
Splice and moving target is not present in video scenery, in the preferable situation of geometrical registration, it is easy to obtain more satisfactory panorama
Image.However, can inevitably have the moving targets such as pedestrian in scene during image collection.There are the dynamics of moving target
In scene, when splicing to image to be spliced, for the panoramic mosaic image for obtaining high quality, splicing line should be made to get around movement mesh
Mark avoids in splicing splicing line from passing through moving target and image is caused to misplace.
There are mainly four types of the joining methods based on dynamic scene in the prior art: first is that using the side of markov random file
Method;Second is that passing through the method for gradient disparities and edge strengthening weight gray difference;Third is that energy-optimised algorithm is cut by figure, in conjunction with
The method of the exposure mask of depth information, smooth transition standard and moving region.However the above method is required to reference to several images,
And the segmentation of moving target is influenced by background, illumination big, when being spliced for two width images, effect is bad.Fourth is that adopting
The moving target in image is extracted with color space method, differential technique, morphology etc., by edge detection operator to movement
Target is matched, and cuts energy-optimised algorithm detection splicing line in conjunction with figure;However, this method wants intensity of illumination and background color
Ask higher, imperfect to moving Object Segmentation when moving target is more, there are splicing lines in splicing wears from moving target
The case where crossing.
Summary of the invention
In order to obtain optimal splicing line when splicing to two width images, splicing line in splicing is avoided to pass through movement
Target, the present invention provide a kind of full-view image joining method and system.
The technical scheme to solve the above technical problems is that
In a first aspect, the embodiment of the invention provides a kind of full-view image joining methods, which comprises
Step 1: by the adjustment of image to be spliced of several sequence arrangements and adjacent mutual overlapping region in same coordinate
System.
Step 2: the image to be spliced described in every width after correction is split, and is obtained in image to be spliced described in every width
All moving targets.
Step 3: motion state judgement being carried out to the moving target, motion state is obtained and determines result.
Step 4: significance detection being carried out to the moving target, determines that the moving target is aobvious compared to image background
Work degree is determined in conjunction with the motion state as a result, assigning corresponding weight, acquisition moving target weight to the moving target
Figure.
Step 5: between image to be spliced described in two width for obtaining mutual overlapping region according to the moving target weight map
Optimal splicing line, according to the optimal splicing line carry out splicing obtain full-view image.
Second aspect, the present invention provides a kind of full-view image splicing system, the system comprises:
Correction module: by the adjustment of image to be spliced of several sequence arrangements and adjacent mutual overlapping region in same seat
Mark system.
Segmentation module: the image to be spliced described in every width after correction is split, and obtains image to be spliced described in every width
Interior all moving targets.
Determination module: carrying out motion state judgement to the moving target, obtains motion state and determines result.
Detection module: significance detection is carried out to the moving target, determines the moving target compared to image background
Significance, determine to obtain moving target power as a result, assign corresponding weight to the moving target in conjunction with the motion state
Multigraph.
Splicing module: image to be spliced described in two width for obtaining mutual overlapping region according to the moving target weight map it
Between optimal splicing line, according to the optimal splicing line carry out splicing obtain full-view image.
The beneficial effect of a kind of full-view image joining method of the invention and system is: being split and obtains to image to be spliced
Moving target is obtained, the motion state of moving target is determined, according to the motion state of moving target and moving target relative to shadow
As the significance of background carries out tax power to moving target, can be accurately identified according to the weight of moving target when obtaining splicing line
The edge of moving target, makes splicing line accurately get around moving target, and splicing line in splicing is avoided to cause across moving target
Image dislocation, the present invention can quickly and accurately obtain the optimal splicing line between the image to be spliced of mutual overlapping region, root
The full-view image that splicing generates clean background, dislocation-free is carried out according to optimal splicing line.
Detailed description of the invention
Fig. 1 is a kind of flow diagram of full-view image joining method of the embodiment of the present invention;
Fig. 2 is that the moving target example of the embodiment of the present invention divides schematic diagram;
Fig. 3 is that the moving Object Segmentation of the embodiment of the present invention merges schematic diagram;
Fig. 4 is that the moving target motion state of the embodiment of the present invention determines schematic diagram;
Fig. 5 is the moving target conspicuousness detection schematic diagram of the embodiment of the present invention;
Fig. 6 is the splicing schematic diagram of two images to be spliced of the embodiment of the present invention;
Fig. 7 is the schematic diagram that energy-optimised algorithm searching splicing line is cut using figure of the embodiment of the present invention;
Fig. 8 is that 360 degree of full-view images of the embodiment of the present invention splice schematic diagram;
Fig. 9 is a kind of full-view image splicing system structural schematic diagram of the embodiment of the present invention.
Specific embodiment
The principle and features of the present invention will be described below with reference to the accompanying drawings, and the given examples are served only to explain the present invention, and
It is non-to be used to limit the scope of the invention.
As shown in Figure 1, a kind of flow diagram of full-view image joining method of the embodiment of the present invention, in the present embodiment with
Pedestrian is described as moving target, which comprises
Step 1: by the adjustment of image to be spliced of several sequence arrangements and adjacent mutual overlapping region in same coordinate
System.
Specifically, the image to be spliced is flake image, using five end to end flake images to be spliced, to obtain
To 360 degree of full-view images.
Step 2: the image to be spliced described in every width after correction is split, and is obtained in image to be spliced described in every width
All moving targets.
Step 3: motion state judgement being carried out to the moving target, motion state is obtained and determines result.
Step 4: significance detection being carried out to the moving target, determines that the moving target is aobvious compared to image background
Work degree is determined in conjunction with the motion state as a result, assigning corresponding weight, acquisition moving target weight to the moving target
Figure.
Step 5: between image to be spliced described in two width for obtaining mutual overlapping region according to the moving target weight map
Optimal splicing line, according to the optimal splicing line carry out splicing obtain full-view image.
In the present embodiment, acquisition moving target is split to image to be spliced, determines the motion state of moving target, root
Tax power is carried out to moving target relative to the significance of image background according to the motion state and moving target of moving target, according to
The weight of moving target can accurately identify the edge of moving target when obtaining splicing line, and splicing line is made accurately to get around movement mesh
Mark avoids in splicing splicing line from passing through moving target and image is caused to misplace, and the present invention can quickly and accurately obtain mutual
Optimal splicing line between the image to be spliced of overlapping region carries out splicing according to optimal splicing line and generates clean background, error-free
The full-view image of position.
Preferably, use PTGUI software by adjustment of image to be spliced in the same coordinate system in step 1.
Preferably, the step 2 specifically includes:
Step 2.1: being the stripping and slicing image of several adjacent mutual overlapping region by the image longitudinal direction to be spliced cutting.
Specifically, it will correct and respectively punctured 1/6 height in the top and the bottom of all images to be spliced of the same coordinate system
Degree gets rid of the image to be spliced top and bottom of not moving target, and the remaining image that will puncture top and the bottom is longitudinal
Cutting is the stripping and slicing image of several adjacent mutual overlapping region, and the degree of overlapping between the stripping and slicing image of mutual overlapping region is
40%.
Moving target example as shown in Figure 2 divides schematic diagram, and Fig. 2 (a) is image to be spliced, and Fig. 2 (b) is by Fig. 2
(a) top and the bottom puncture the remaining image after 1/6 height, Fig. 2 (c) be will obtain after the longitudinal cutting of Fig. 2 (b) adjacent mutually
There is the stripping and slicing image of overlapping region.
Step 2.2: the stripping and slicing image being split using the example segmentation network based on deep learning, is obtained described
The position coordinates of the moving target and the moving target in the image to be spliced in stripping and slicing image.
Specifically, stripping and slicing image is inputted into pre-set MNC (Multi-task Network Cascades) depth
Learn example and divide network model, is distinguished by example, exposure mask estimation, the processing of three subtasks of target classification, obtain stripping and slicing
The position coordinates of all moving targets and moving target based on image to be spliced in image form cascade between three subtasks
Structure shares convolution feature.
The segmentation of target level example is carried out to moving target using the example segmentation network based on deep learning, can solve movement
The problems such as Objective extraction is imperfect, multiple moving targets are difficult to complete parttion, guarantees the integrality of moving target being partitioned into.
Preferably, after being split to the overlapping region of the stripping and slicing image, the repeating motion target being partitioned into is merged
At a moving target.It prevents erroneous detection and repeats to detect, improve segmentation precision.
Moving Object Segmentation as shown in Figure 3 merges schematic diagram, and Fig. 3 (a) is a width stripping and slicing image, and Fig. 3 (b) is and Fig. 3
(a) the stripping and slicing image of mutual overlapping region, moving target is the same fortune in the overlapping region of Fig. 3 (a) and Fig. 3 (b) in box
This repeating motion target being partitioned into is merged into a moving target in Fig. 3 (c) by moving-target.
Preferably, the step 3 specifically includes:
Step 3.1: right according to the position coordinates of the moving target and the moving target in the image to be spliced
The image to be spliced in any two width mutual overlapping region, by image to be spliced described in a width the moving target and
The moving target in image to be spliced described in another width is matched.
Specifically, in the present embodiment using pedestrian's weight identification technology between in two images to be spliced of mutual overlapping region
Pedestrian target matched, bilateral symmetry is based on to the head, trunk, leg of human body according to pedestrian body symmetric characteristics first
Axis is divided, and is extracted each region and is added up the features such as color, statistic texture, and is weighted based on symmetrical axis to feature,
Closer to the feature of axis, its weight is higher, and the feature weight far from axis is lower, is matched according to the weight of acquisition.
Step 3.2: motion state being carried out to the moving target and is sentenced according to scheduled criterion according to matching result
It is fixed.
Preferably, the scheduled criterion is indicated by the first formula, first formula are as follows:
M (i, j)=p (i) * exp (r (i, j))
Wherein, i and j respectively represents the moving target in the image to be spliced of two width mutual overlapping regions, m
(i, j) represents the motion state of moving target i and moving target j.P (i) represents the matching knot of moving target i Yu moving target j
Fruit, if successful match, value 1;Otherwise value is 0.R (i, j) represents the moving target i and moving target j of successful match
Minimum bounding box degree of overlapping, the minimum bounding box degree of overlapping indicates by the second formula, second formula are as follows:
R (i, j)=o (i, j)/(a (i)+a (j)-o (i, j))
Wherein, o (i, j) represents the overlapping area between moving target i and moving target j, and a (i) and a (j) are respectively represented
The minimum bounding box area of moving target i and moving target j.Bounding box refers to the minimum that can completely include some object
Cube.
Specifically, as m (i, j)=0, moving target i is judged to disappearing;As m (i, j)=1, moving target i and movement
Target j is determined as grand movement;As m (i, j) ∈ (1, e), moving target i and moving target j are determined as that small range moves;It is no
Then, moving target i and moving target j is judged to not moving.
Moving target motion state as shown in Figure 4 determines that schematic diagram, Fig. 4 (a) are a certain left image to be spliced, Fig. 4
(d) it is the right image to be spliced with Fig. 4 (a) mutual overlapping region, is the overlay region after Fig. 4 (a) and Fig. 4 (d) correction in box
Domain, Fig. 4 (b) are all moving targets in Fig. 4 (a), and Fig. 4 (e) is all moving targets in Fig. 4 (d), will be in Fig. 4 (b)
Moving target is matched with the moving target in Fig. 4 (e) respectively, is the moving target of successful match, movement in Fig. 4 (c)
State is judged as grand movement, is the disappearance moving target of non-successful match in Fig. 4 (f).
Preferably, the step 4 specifically includes:
Step 4.1: according to HC (Histogram-based Contrast) conspicuousness detection method, determining the movement mesh
Mark the significance compared to image background.
Specifically, color contrast algorithm of the HC conspicuousness detection method based on global color histogram uses movement mesh
The color contrast difference of every other pixel determines that its significant angle value, difference are got in each pixel of target and image background
Big significant angle value is bigger.
Moving target conspicuousness detection schematic diagram as shown in Figure 5, Fig. 5 (a) are a certain image to be spliced, are in box
Moving target in image to be spliced, Fig. 5 (b) are the moving target saliency map obtained after the detection of HC conspicuousness.
Step 4.2: according to the significance, the motion state obtained in conjunction with step 3 determines result to the movement
Target assigns weight, obtains the moving target weight map of the image to be spliced.
Preferably, the motion state determines that result moves and do not move four kinds including disappearance, grand movement, small range
State, the motion state that the combination step 3 obtains determine that result assigns the specific implementation of weight to the moving target
Are as follows:
Enabling Img1 and Img2 is the image to be spliced of any two width mutual overlapping region, and i and j are respectively shadow to be spliced
As the moving target in Img1 and image Img2 to be spliced.
For the moving target i in image Img1 to be spliced, determined according to the motion state as a result, if moving target i
It disappears, there is no the movements to match with the moving target i in the image Img2 to be spliced with Img1 mutual overlapping region
Target then assigns weight W to moving target i1(i)=θ.
If there is the moving target j to match with moving target i in image Img2 to be spliced, when moving target i and movement
When target j is determined as grand movement, moving target i and moving target j is subjected to significance comparison, H (Ok) represent it is any to
The conspicuousness of any moving target k in splicing image Img is finally measured, and the conspicuousness is finally measured by third formula table
Show, the third formula are as follows:
Wherein, OkRegion belonging to referring to moving target k in image Img to be spliced, is defined as motion target area, HC
(x) refer to positioned at motion target area OkThe single pixel saliency value that interior pixel x is obtained based on HC algorithm, cnt (Ok) refer to movement mesh
Mark region OkInterior number of pixels is finally measured according to the conspicuousness that third formula obtains moving target i and moving target j respectively
H(Oi) and H (Oj), OiIt is the moving target i motion target area affiliated in image Img1 to be spliced, OiIt is movement mesh
Mark the j motion target area affiliated in image Img2 to be spliced.
To H (Oi) and H (Oj) be normalized, obtain normalized value T (Oi) and T (Oj), the weight W of moving target i1(Oi)
With the weight W of moving target j2(Oj) indicated by the 4th formula, the 4th formula are as follows:
Wherein θ is selected weighted value, is set as 255.
When the moving target i and moving target j that match are determined as small range movement, by moving target i and movement mesh
It marks j and merges imparting weight W1(Oi∪Oj)=θ, W2(Oi∪Oj)=θ.
When the moving target i and moving target j that match are judged to not moving, to moving target i and moving target j points
It Fu Yu not weight W1(Oi)=0, W2(Oj)=0.
Simultaneously set the weight of remaining non-athletic target area of image Img1 to be spliced and image Img2 to be spliced to
0, remaining non-athletic target area refers in image Img1 to be spliced and image Img2 to be spliced other than motion target area
All areas.
The moving target weight map that result generates every image to be spliced is weighed according to above-mentioned tax, i.e. moving target cum rights is covered
Film, for extracting the motion target area in image to be spliced.
The splicing schematic diagram of two images to be spliced as shown in FIG. 6, Fig. 6 (a) and Fig. 6 (c) are respectively mutual overlay region
The left image to be spliced and right image to be spliced in domain, the interior overlapping region for two images to be spliced of box, Fig. 6 (b) and Fig. 6
(d) be respectively overlapping region in Fig. 6 (a) and Fig. 6 (c) moving target weight map, Fig. 6 (e) is that Fig. 6 (a) and Fig. 6 (c) splices
The image obtained afterwards.
Preferably, the specific implementation of step 5 are as follows:
Energy-optimised algorithm is cut using figure, according to the moving target weight map, obtain mutual overlapping region it is described to
Splice the optimal splicing line between image, spliced according to the optimal splicing line, obtains full-view image.
Specifically, energy-optimised calculation is cut using figure for two image Img1 and Img2 to be spliced of mutual overlapping region
Method finally splices the energy term E (I) of obtained full-view image I by data capacity item Edata(I) and smooth energy term Esmooth(I)
Composition, energy term E (I) shows as the total power consumption that splicing line is passed through from image in image joint, obtains optimal splicing
The process of line is exactly to find the process of least energy item E (I), and the energy term E (I) is indicated by the 5th formula, and the described 5th is public
Formula are as follows:
E (I)=Edata(I)+Esmooth(I)
Wherein, data capacity item Edata(I) it is indicated by the 6th formula, the 6th formula are as follows:
WithIt respectively represents pixel p in full-view image I and comes from image Img1 to be spliced and image to be spliced
The energy consumption of Img2.Motion target area is extracted according to moving target weight map, for the moving target in image to be spliced
Region and non-athletic target area, non-athletic target area refer to all areas in image to be spliced other than motion target area
Domain,WithCalculation it is different.
For the motion target area O of image Img1 to be spliced and image Img2 to be splicediAnd Oj,With
Not by the weight W of pixel p motion target area in image Img1 to be spliced and image Img2 to be spliced1(Oi) and W2(Oj) certainly
It is fixed, energy consumption of the pixel p from motion target areaWithIt is indicated by the 7th formula, the 7th formula
Are as follows:
T represents the penalty coefficient of equilibrium data energy term and smooth energy term, and the value of T is set as 100 in the present embodiment.
For the non-athletic target area of image Img1 to be spliced and image Img2 to be spliced, if pixel p is to be spliced
Image Img1 then willValue be set as 0,Value be set as infinitely great;If pixel p comes from image Img2 to be spliced,
Then willValue be set as infinitely great,Value be set as 0.Energy consumption of the pixel p from non-athletic target areaWithIt is indicated by the 8th formula, the 8th formula are as follows:
By the 8th formula it is found that only when pixel p is simultaneously from image Img1 to be spliced and image Img2 to be spliced,
Data capacity item EdataIt (I) is 0, otherwise, data capacity item EdataIt (I) is infinity.To keep energy term E (I) minimum, define
Splicing line is passed through from the overlapping region of image Img1 to be spliced and image Img2 to be spliced.
Smoothed energy item Esmooth(I) energy consumption of all pixels in neighborhood is represented, is indicated by the 9th formula, it is described
9th formula are as follows:
Wherein, N (I) represents all pixels pair after splicing in full-view image I, and (p, q) represents adjacent pixel, factor sigma (p,
Q) it represents pixel p and whether q is identical, then σ (p, q) is set as 0 if they are the same, is otherwise set as 1. Esmooth(p, q) represents pixel p and q
Between smoothed energy item, the design summation of smoothed energy item considers gray scale, gradient difference, Texture complication, thus non-athletic
Target area guarantees that splicing line is passed through from the low region of uniform gray level, Texture complication as far as possible.
Gray scale, gradient difference and Texture complication can be calculated by the prior art.
(Graph-Cuts) energy optimizing method is cut using figure, to the data capacity item E of overlapping area imagedata(I)、
Smoothed energy item Esmooth(I) it is optimized with energy term E (I), optimization method includes α-expansion algorithm and alpha-beta-swap
Algorithm, entire algorithm loop iteration no longer reduce up to energy term E (I) is minimum, obtain optimal splicing line, be implemented as existing
Technology.
As shown in Figure 7 cuts the rough schematic that energy-optimised algorithm finds splicing line using figure, and square is two width in figure
The pixel of image overlap area to be spliced, S and T are terminal vertex, and there are a lines between every two pixel, its weight is by upper
The decision of smoothed energy item is stated, the thickness on side represents the size of smoothed energy, and the more thick corresponding energy in side is bigger;Pixel and terminal top
The weight on side is determined that energy term represents splicing line from the total power consumption passed through between pixel by above-mentioned data capacity item between point.
The energy minimum cut according to figure is split, and obtains optimal splicing line, cut is optimal splicing line in figure.
The imagery zone that splicing panorama image is used in image to be spliced is obtained according to optimal splicing line, by the image area
Domain carries out splicing and obtains full-view image.
360 degree of full-view images as shown in Figure 8 splice schematic diagram, and Fig. 8 (a) is the raw video of image to be spliced, Fig. 8
It (b) is final full-view image splicing result.
As shown in figure 9, the present invention provides a kind of full-view image splicing system, the system comprises:
Correction module: by the adjustment of image to be spliced of several sequence arrangements and adjacent mutual overlapping region in same seat
Mark system.
Segmentation module: the image to be spliced described in every width after correction is split, and obtains image to be spliced described in every width
Interior all moving targets.
Determination module: carrying out motion state judgement to the moving target, obtains motion state and determines result.
Detection module: significance detection is carried out to the moving target, determines the moving target compared to image background
Significance, determine to obtain moving target power as a result, assign corresponding weight to the moving target in conjunction with the motion state
Multigraph.
Splicing module: image to be spliced described in two width for obtaining mutual overlapping region according to the moving target weight map it
Between optimal splicing line, according to the optimal splicing line carry out splicing obtain full-view image.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of full-view image joining method, which is characterized in that the described method includes:
Step 1: by the adjustment of image to be spliced of several sequence arrangements and adjacent mutual overlapping region in the same coordinate system;
Step 2: the image to be spliced described in every width after correction is split, and is obtained all in image to be spliced described in every width
Moving target;
Step 3: motion state judgement being carried out to the moving target, motion state is obtained and determines result;
Step 4: significance detection being carried out to the moving target, determines that the moving target is significant compared to image background
Degree is determined in conjunction with the motion state as a result, assigning corresponding weight, acquisition moving target weight map to the moving target;
Step 5: between image to be spliced described in two width for obtaining mutual overlapping region according to the moving target weight map most
Excellent splicing line carries out splicing according to the optimal splicing line and obtains full-view image.
2. full-view image joining method according to claim 1, which is characterized in that the step 2 specifically includes:
Step 2.1: being the stripping and slicing image of several adjacent mutual overlapping region by the image longitudinal direction to be spliced cutting;
Step 2.2: the stripping and slicing image being split using the example segmentation network based on deep learning, obtains the stripping and slicing
The position coordinates of the moving target and the moving target in the image to be spliced in image.
3. full-view image joining method according to claim 2, which is characterized in that described to use the reality based on deep learning
The specific implementation that example segmentation network is split the stripping and slicing image are as follows:
After being split to the overlapping region of the stripping and slicing image, the repeating motion target being partitioned into is merged into a movement mesh
Mark.
4. full-view image joining method according to claim 3, which is characterized in that the step 3 specifically includes:
Step 3.1: according to the position coordinates of the moving target and the moving target in the image to be spliced, for appointing
It anticipates the images to be spliced of two width mutual overlapping regions, by the moving target in image to be spliced described in a width and another
The moving target in image to be spliced described in width is matched;
Step 3.2: according to matching result, according to scheduled criterion, motion state judgement being carried out to the moving target.
5. full-view image joining method according to claim 4, which is characterized in that the scheduled criterion is by first
Formula expression, first formula are as follows:
M (i, j)=p (i) * exp (r (i, j))
Wherein, i and j respectively represents the moving target in image to be spliced described in two width of mutual overlapping region, m (i, j)
The motion state of the moving target i and the moving target j are represented, p (i) represents the moving target i and the movement mesh
The matching result for marking j, if successful match, value 1, otherwise value is the movement mesh that 0, r (i, j) represents successful match
The minimum bounding box degree of overlapping of i and the moving target j are marked, the minimum bounding box degree of overlapping is indicated by the second formula, described
Second formula are as follows:
R (i, j)=o (i, j)/(a (i)+a (j)-o (i, j))
Wherein, o (i, j) represents the overlapping area between the moving target i and the moving target j, and a (i) and a (j) are respectively
The minimum bounding box area of the moving target i and the moving target j are represented, bounding box, which is one, can completely include some
The minimum cube of object.
6. full-view image joining method according to claim 1, which is characterized in that the step 4 specifically includes:
Step 4.1: according to HC conspicuousness detection method, determining significance of the moving target compared to the image background;
Step 4.2: according to the significance, the motion state obtained in conjunction with step 3 determines result to the moving target
Weight is assigned, the moving target weight map of the image to be spliced is obtained.
7. full-view image joining method according to claim 6, which is characterized in that the motion state determines that result includes
Disappearance, grand movement, small range move and do not move four kinds of states, and the motion state that the combination step 3 obtains is sentenced
Determine the specific implementation that result assigns weight to the moving target are as follows:
Enabling Img1 and Img2 is the image to be spliced of any two width mutual overlapping region, and i and j are respectively image to be spliced
The moving target in Img1 and image Img2 to be spliced;
For the moving target i in the image Img1 to be spliced, determined according to the motion state as a result, if the fortune
Moving-target i is judged to disappearing, in the image Img2 to be spliced of the image Img1 mutual overlapping region to be spliced not
In the presence of the moving target to match with the moving target i, then weight W is assigned to the moving target i1(i)=θ;
If there is the moving target j to match with the moving target i in the image Img2 to be spliced, when the movement
When target i and the moving target j are determined as grand movement, the moving target i and moving target j is carried out significant
Degree compares, H (Ok) conspicuousness that represents any moving target k in any image Img to be spliced finally measures, the conspicuousness
Finally measurement is indicated by third formula, the third formula are as follows:
Wherein, OkRegion belonging to referring to the moving target k in the image Img to be spliced, is defined as motion target area,
HC (x) refers to positioned at the motion target area OkThe single pixel saliency value that interior pixel x is obtained based on HC algorithm, cnt (Ok) refer to
The motion target area OkInterior number of pixels obtains the moving target i and the fortune according to the third formula respectively
The conspicuousness of moving-target j finally measures H (Oi) and H (Oj), OiIt is moving target i institute in the image Img1 to be spliced
The motion target area belonged to, OiIt is the moving target j movement mesh affiliated in the image Img2 to be spliced
Mark region;
To the H (Oi) and the H (Oj) be normalized, obtain normalized value T (Oi) and T (Oj), the power of the moving target i
Value W1(Oi) and the moving target j weight W2(Oj) indicated by the 4th formula, the 4th formula are as follows:
Wherein, θ is the weighted value of setting;
When the moving target i and the moving target j that match are determined as small range movement, by the moving target i
Merge with the moving target j and assigns weight W1(Oi∪Oj)=θ, W2(Oi∪Oj)=θ;
When the moving target i and the moving target j that match are judged to not moving, to the moving target i and institute
It states moving target j and assigns weight W respectively1(Oi)=0, W2(Oj)=0;
The weight of remaining non-athletic target area of the image Img1 to be spliced and the image Img2 to be spliced is set simultaneously
It is set to 0.
8. full-view image joining method according to claim 1-7, which is characterized in that the step 5 it is specific
It realizes are as follows:
Energy-optimised algorithm is cut using figure, according to the moving target weight map, obtains the described to be spliced of mutual overlapping region
Optimal splicing line between image is spliced according to the optimal splicing line, obtains full-view image.
9. full-view image joining method according to claim 1-7, which is characterized in that the step 1 it is specific
It realizes are as follows: using PTGUI software by the adjustment of image to be spliced in the same coordinate system.
10. a kind of full-view image splicing system, which is characterized in that the system comprises:
Correction module: by the adjustment of image to be spliced of several sequence arrangements and adjacent mutual overlapping region in the same coordinate system;
Segmentation module: the image to be spliced described in every width after correction is split, and is obtained in image to be spliced described in every width
All moving targets;
Determination module: carrying out motion state judgement to the moving target, obtains motion state and determines result;
Detection module: carrying out significance detection to the moving target, determines that the moving target is aobvious compared to image background
Work degree is determined in conjunction with the motion state as a result, assigning corresponding weight, acquisition moving target weight to the moving target
Figure;
Splicing module: between image to be spliced described in two width for obtaining mutual overlapping region according to the moving target weight map
Optimal splicing line carries out splicing according to the optimal splicing line and obtains full-view image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810811294.4A CN109146782A (en) | 2018-07-23 | 2018-07-23 | A kind of full-view image joining method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810811294.4A CN109146782A (en) | 2018-07-23 | 2018-07-23 | A kind of full-view image joining method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109146782A true CN109146782A (en) | 2019-01-04 |
Family
ID=64801349
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810811294.4A Withdrawn CN109146782A (en) | 2018-07-23 | 2018-07-23 | A kind of full-view image joining method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109146782A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111312403A (en) * | 2020-01-21 | 2020-06-19 | 山东师范大学 | Disease prediction system, device and medium based on instance and feature sharing cascade |
CN112184541A (en) * | 2019-07-05 | 2021-01-05 | 杭州海康威视数字技术股份有限公司 | Image splicing method, device and equipment and storage medium |
CN113344957A (en) * | 2021-07-19 | 2021-09-03 | 北京城市网邻信息技术有限公司 | Image processing method, image processing apparatus, and non-transitory storage medium |
-
2018
- 2018-07-23 CN CN201810811294.4A patent/CN109146782A/en not_active Withdrawn
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112184541A (en) * | 2019-07-05 | 2021-01-05 | 杭州海康威视数字技术股份有限公司 | Image splicing method, device and equipment and storage medium |
CN111312403A (en) * | 2020-01-21 | 2020-06-19 | 山东师范大学 | Disease prediction system, device and medium based on instance and feature sharing cascade |
CN113344957A (en) * | 2021-07-19 | 2021-09-03 | 北京城市网邻信息技术有限公司 | Image processing method, image processing apparatus, and non-transitory storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109615611B (en) | Inspection image-based insulator self-explosion defect detection method | |
CN111899334B (en) | Visual synchronous positioning and map building method and device based on point-line characteristics | |
Shin et al. | Vision-based navigation of an unmanned surface vehicle with object detection and tracking abilities | |
CN108304798B (en) | Street level order event video detection method based on deep learning and motion consistency | |
Coughlan et al. | The manhattan world assumption: Regularities in scene statistics which enable bayesian inference | |
US9245345B2 (en) | Device for generating three dimensional feature data, method for generating three-dimensional feature data, and recording medium on which program for generating three-dimensional feature data is recorded | |
CN109961399B (en) | Optimal suture line searching method based on image distance transformation | |
CN110473221B (en) | Automatic target object scanning system and method | |
WO2010017255A2 (en) | Cut-line steering methods for forming a mosaic image of a geographical area | |
Urban et al. | Finding a good feature detector-descriptor combination for the 2D keypoint-based registration of TLS point clouds | |
CN109146782A (en) | A kind of full-view image joining method and system | |
CN108681711A (en) | A kind of natural landmark extracting method towards mobile robot | |
CN117036641A (en) | Road scene three-dimensional reconstruction and defect detection method based on binocular vision | |
CN113744315B (en) | Semi-direct vision odometer based on binocular vision | |
Cherian et al. | Accurate 3D ground plane estimation from a single image | |
Xiao et al. | Geo-spatial aerial video processing for scene understanding and object tracking | |
CN112347967B (en) | Pedestrian detection method fusing motion information in complex scene | |
CN110717910B (en) | CT image target detection method based on convolutional neural network and CT scanner | |
CN113096016A (en) | Low-altitude aerial image splicing method and system | |
Haines et al. | Estimating Planar Structure in Single Images by Learning from Examples. | |
CN116958837A (en) | Municipal facilities fault detection system based on unmanned aerial vehicle | |
D’Amicantonio et al. | Homography estimation for camera calibration in complex topological scenes | |
Mecocci et al. | Outdoor scenes interpretation suitable for blind people navigation | |
Park et al. | Line-based single view 3D reconstruction in Manhattan world for augmented reality | |
CN117223034A (en) | Method and apparatus for generating an countermeasure patch |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190104 |
|
WW01 | Invention patent application withdrawn after publication |