CN105894068A - FPAR card design method and rapid identification and positioning method of FPAR card - Google Patents
FPAR card design method and rapid identification and positioning method of FPAR card Download PDFInfo
- Publication number
- CN105894068A CN105894068A CN201610137618.1A CN201610137618A CN105894068A CN 105894068 A CN105894068 A CN 105894068A CN 201610137618 A CN201610137618 A CN 201610137618A CN 105894068 A CN105894068 A CN 105894068A
- Authority
- CN
- China
- Prior art keywords
- fpar
- card
- point
- concentric
- anchor point
- 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
- G06K—GRAPHICAL DATA READING; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
- G06K19/00—Record carriers for use with machines and with at least a part designed to carry digital markings
- G06K19/06—Record carriers for use with machines and with at least a part designed to carry digital markings characterised by the kind of the digital marking, e.g. shape, nature, code
- G06K19/06009—Record carriers for use with machines and with at least a part designed to carry digital markings characterised by the kind of the digital marking, e.g. shape, nature, code with optically detectable marking
- G06K19/06046—Constructional details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Computer Graphics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a fast positioning augmented reality (FPAR) card design method and a rapid identification and positioning method of an FPAR card. The design method is characterized in that a plurality of groups of concentric circles/concentric polygonal lines are arranged on an FPAR card; each group of concentric circles/concentric polygonal lines is called as a positioning point; the groups of concentric circles/concentric polygonal lines form an identification line pattern; and a general pattern after arrangement of the groups of concentric circles/concentric polygonal lines is in a central symmetric mode. In addition, the rapid identification and positioning method comprises: S1, video inputting is carried out; and a collected video source file is decomposed into a frame sequence; S2, RGB/YUV pretreatment is carried out; S3, binarization processing is carried out; to be specific, binary separation is carried out on an image by using a blurring filter method; S4, image identification processing is carried out by using a color filling method; S5, a possible positioning point is determined; S6, with an imaging picture, a normal vector of the FPAR card is obtained by using an inverse method, so that subsequent virtual reality rendering can be carried out conveniently; S7, the position of the overall FPAR card in three-dimensional space is determined; and S8, virtual reality rendering is carried out. The methods have advantages of good zooming, jittering, blurred image identification, partial shield identification, rotation performances, high robustness and reliability, high static accuracy, and fast identification speed.
Description
Technical field
The present invention relates to the research field of image procossing, particularly to a kind of FPAR card design and quick recognition positioning method.
Background technology
Augmented reality (Augmented Reality, it is called for short AR) technology, it is a kind of by new technique integrated to real world information and virtual world information " seamless ", it is the entity information (visual information originally being difficult to experience in the certain time spatial dimension of real world, sound, taste, sense of touch etc.), by science and technology such as computers, superposition again after analog simulation, virtual information is applied to real world, by the perception of human sensory institute, thus reaches the sensory experience of exceeding reality.Real environment and dummy object have been added to same picture in real time or space exists simultaneously.
Augmented reality contains multimedia, three-dimensional modeling, real-time video show and control, Multi-sensor Fusion, real-time tracking and new technique and the means such as registration, scene fusion.Augmented reality provides in the ordinary course of things, is different from the appreciable information of the mankind.
FPAR (Fast Positioning Augmented Reality), FPAR are to integrate quickly location and the AR card of quick identification information.
Summary of the invention
Present invention is primarily targeted at the shortcoming overcoming prior art with not enough, it is provided that a kind of FPAR card design and quick recognition positioning method.
In order to achieve the above object, the present invention is by the following technical solutions:
The invention provides the method for designing of a kind of FPAR card, described FPAR card is provided with organizes concentric circular/concentric polygon lines more, each group of concentric circular/concentric polygon is referred to as an anchor point, organize concentric circular/concentric polygon composition more and identify grain pattern, described organize concentric circular/concentric polygon arrangement after total figure be centrosymmetric.
As preferred technical scheme, described FPAR card is provided with 5 groups of concentric circulars/concentric polygon lines, the lines shape often organizing anchor point is identical, wherein 4 groups of concentric circular/concentric polygon are arranged in the corner of rectangular card, and the 5th group of concentric circular/concentric polygon is scheduled on the middle position on the long limit of card or is on the asymmetric position being capable of identify that FPAR card；Often the scaling at group group concentric circular/concentric polygon two adjacent rings edge is
Present invention also offers the quick recognition positioning method of a kind of FPAR card, comprise the steps:
S1, video input, first resolve into frame sequence by the video source file collected；
S2, RGB/YUV pretreatment, it is RGB or yuv format that the frame extracted is probably, when extract for rgb format time, method for quickly identifying is only to take redness or green channel, and rest channels is abandoned, when extract for yuv format time, as long as then Y passage, abandon U, V passage；
S3, binary conversion treatment, carry out two-value separation to image；
S4, employing the Blur filter method carry out image recognition processing；Using the double iir filter of positive inverted order, wherein forward is P0=p0, other Pn=w0*pn+w1*Pn-1+w2*Pn-2, similarly, reverse form is Qlen-1=Plen-1, remaining Qn=w0*Pn+w1*Qn+1+w2*Qn+2, wherein the part of subscript out of bounds uses the mode of clamper to process, and i.e. takes the legal subscript of its arest neighbors, uses iir filter to emulate Gaussian Blur with such form；
S5, judge possible anchor point,
S6, the 3D approach obtaining FPAR card are vectorial, by the anti-normal vector pushing away FPAR card of imaging picture, render serving follow-up virtual reality；
S7, by comparison and drift correction, determine the three-dimensional position of each anchor point, so that it is determined that whole FPAR is stuck in three-dimensional position；
S8, rendering virtual reality, be attached to by a sheet of planar picture virtually on three-dimensional FPAR card, calculate after picture is attached on FPAR card, the color of each corresponding point should show that position on imaging plane.
As preferred technical scheme, in described step S3, realize binary conversion treatment by the Blur filter, itself particularly as follows:
By two copies of image copy, it is designated as copy 1 and copy 2；
If there being more apparent noise in video, copy 1 carries out the Gaussian Blur of extremely trace, wherein σ≤3px；
Copy 2 carries out the Gaussian Blur that radius is bigger, and radius is 1/30 to the 1/10 of image diagonal；
Setting up a blank dot matrix image, size is consistent with front two copies, is designated as M, place brighter than copy 2 for correspondence position copy 1 is set to 1, is otherwise set to 0, then M is result.
As preferred technical scheme, in described step S4, carry out image procossing method particularly includes:
S41, iir filter by simulation Gaussian Blur determine w0, w1, w2 these three floating point values, and P and p is the numerical value between 0-255, and wherein P is floating-point, and p is integer；Therefore the result of product of part P and p Yu w is carried out pre-stored, make a table, with use for future reference；
S42, the way of a kind of approximation is used to list a less precomputation multiplication table；
Approximate formula is P*w ≈ P ' * W=W [P '], and p*w=W ' [p], wherein W and W ' is the precalculated result of product of storage of array, and P ' is the result after single precision floating datum P blocks 16 mantissa bits.
As preferred technical scheme, in described step S4, addition of integer is used to replace floating-point addition operation, in IIR positive sequence direction, owing to comprising fractional part in calculating, therefore employ integer calculations and calculate to simulate fractional fixed point, carry out integer calculations with 16 shorts, wherein integer and fractional part respectively account for 8, p* ω * 256=(p ＜ ＜ 8) * ω ≈ [p* ω * 256]=Ω ' [p]；The computing formula of the IIR of P* ω ≈ [p* ω] ≈ Ω [p '], the most actually positive sequence is Pn=ω0*(pn<<8)+ω1*Pn-1+ω2*Pn-2, PnFor integer, result of calculation is about 256 times of former formula, and during calculating, available Ω table short-cut multiplication computing, above formula becomes Pn=Ω ' [pn]+Ω[Pn-1]+Ω[Pn-2]；
The scale-up problem of positive numerical sequence can not process, and after inverted order has calculated, directly use moves to right 8 and can repair, and the table that inverted order uses with positive sequence is consistent, simply computing formula slightly difference, for Qn=ω0*Pn+ω 1*Qn+1+ω2*Qn+2, multiplication herein again may be by looking into Ω table and solves, and expression formula is Qn=Ω [Pn]+Ω[Qn-1]+Ω[Qn-2], then Rn=Qn> > 8 it is required.
As preferred technical scheme, step S5 particularly as follows:
S5.1, check each pixel of M, if current pixel is 0 or 1, then with numerical value 2, carry out 4 connections or 8 connections are filled, and add up this Breadth Maximum filled and height, also have the area filled；
S5.2, to fill region limit；Even from (x, y) starts to fill, set maximum and fill region u, then left margin is x-u, and right margin is x+u, and coboundary is y-u, and lower boundary is y+u；So can limit total filling area less than 4u2, thus alleviate internal memory burden；
S5.3, selected possible locating point position, can limit and be legal between Aspect Ratio is at 1:4 to 4:1, and the design additionally, due to anchor point is graftal, so either which laminated striation sample, its filling rate is all consistent, and filling rate is near 0.436；The point only meeting above-mentioned requirements just counts form L；
S5.4, eliminating flase drop point, compare to the item in form, if some frame is independent, or other any frame of getting along well is concentric, then may be considered flase drop；Being merged by the concentric frame that selects of each group in form L, record its average central and mark stores to table H, its mid score needs to consider to relate to this concentricity organizing each frame, length-width ratio concordance, frame overlap number and each select frame filling rate, more meet the requirements, mark is the highest, otherwise the lowest；
S5.5, from anchor point to FPAR card grain pattern identification, the method using permutation and combination, enumerate all of anchor point and substitute into computing formula, the most reasonable to test which kind of corresponded manner.
As preferred technical scheme, in step S5.4, for anchor point identification, for avoiding flase drop, typically will at least 2 layers.
As preferred technical scheme, step S5 particularly as follows: each anchor point judgement particularly as follows:
(1) convex polygon is first judged
Judge that convex polygon method is as follows:
B. get rid of line segment and have intersection option, it is judged that product jack per line；
A, b, c, d, and e are regarded as three-dimensional point, wherein z-axis=0, then calculates Value, these values whether with axle jack per line, and can might as well all be set to positive sign；
(2) Comprehensive Evaluation；
Owing to each anchor point is by image recognition, during operation, it not directly to evaluate whether to meet the requirements, but it is how many for evaluating the laminating degree required with rule；
(3) Optimal Decision-making；
When operational capability is permitted, appointing a little in H, can be chosen 5 and carry out fully intermeshing, substitute into above rule and verify, if the candidate items in H is too many, fully intermeshing is an astronomical figure, and in order to alleviate operand, front n the point that can take H mid score higher carries out fully intermeshing substitution test.
As preferred technical scheme, step S7 particularly as follows: use drawTriangles method, not exclusively calculate on FPAR card color a little, but do a sparse matrix grid, then the conversion of the position triangular fragments in the hole of grid in this mesh mapping to imaging plane is simulated.
The present invention compared with prior art, has the advantage that and beneficial effect:
1, the present invention possesses scaling advantage；
The grain pattern of FPAR card is fractal structure, after removing the black circle of outmost turns, black for inner ring circle is amplified, will obtain artwork；Inverse printing does not affect recognition effect.Such grain pattern has scaling advantage.When the image resolution ratio that photographic head captures is relatively low or FPAR card is distant, bigger outer ring still can be identified；When resolution of lens is the highest or FPAR card is close together, the area of outer ring is excessive, may be not easy to identify, now can ignore outer ring beyond outside camera lens, only identify the inner ring that area is less.
2, the present invention possesses shake and fuzzy image recognition advantage；
Owing to there being the shake of camera lens and scenery, the image that photographic head photographs, it is impossible to the most clearly, the image obtained under many circumstances often produces motion blur.Relatively slightly bigger outer ring, can help to identify anchor point from fuzzy image.
3, static precision；
When photographing FPAR card clearly, being the most singly outer ring, the ringlet at anchor point center also can be analyzed clear, and this contributes to obtaining the higher recognition result of accuracy.
4, advantage is rotated；
Due to the design of concentric circular/concentric polygon, in any case so rotating, the shape of anchor point all as.Do not rotated and the shadow of angle.
5, partial occlusion identification advantage；
As long as the visible region of each anchor point is no less than two concentric circular/concentric polygon.
6, strong robustness, highly reliable；
Either uneven illumination is even, picture noise serious, card tilts, it is serious to rotate, picture is brighter, dark, fuzzy etc. can normally identify.
7, recognition speed is fast.
Compared with similar technique, have and identify faster and detent edge.
Accompanying drawing explanation
Fig. 1 is the schematic diagram of FPAR card of the present invention；
Fig. 2 is the concentric circular scaling schematic diagram of FPAR card of the present invention；
Fig. 3 is the FPAR card recognition effect under more satisfactory environment；
Fig. 4 is the FPAR card recognition effect figure under ambisexuality environment；
Fig. 5 is to block the FPAR card recognition effect figure under environment；
Fig. 6 is the recognition effect figure of the environment that traditional method is low in contrast, noise is many；
Fig. 7 is the design sketch that the present invention processes FPAR card；
Fig. 8 is the FPAR card identification figure under fringe；
Fig. 9 is the binary conversion treatment figure of fixed threshold；
The situation that the colors in schematic diagram of Figure 10 present invention；
Figure 11 is that all frames after the present invention fills select result；
Figure 12 is that the center of grain pattern is by the schematic diagram in repeat block；
Figure 13 is the pattern schematic diagram of possible flase drop；
Figure 14 is the anchor point flase drop pattern constructed；
Figure 15 is the schematic diagram that the pattern of mahjong bobbin may cause flase drop；
Figure 16 region limits method each step schematic diagram；
Figure 17 is that the frame of region limits method selects result figure；
Figure 18 is not use the frame of region limits to select result figure；
Figure 19 is imaging schematic diagram；
Figure 20 is FPAR card plane graph；
Figure 21 is to judge line segment crossing instances；
Figure 22 is the imaging schematic diagrams solved more；
Figure 23 is multiple FPAR cards suspicious anchor point schematic diagrams；
Figure 24 is modified version FPAR card grain pattern figure；
Figure 25 is multiple FPAR card schematic diagrams；
Figure 26 is the FPAR card schematic diagram of band opening；
Figure 27 is that multiple FPAR cards in band direction identify schematic diagram；
Figure 28 is the FPAR card multi-angle distortion image of 3-D view；
Figure 29 is 3-D view shaping schematic view；
Figure 30 is the real-time rendering figure mixed the spurious with the genuine；
Figure 31 is to render sectional drawing；
Figure 32 be FPAR card render extension effect.
Detailed description of the invention
Below in conjunction with embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited to this.
Embodiment
1.AR technology introduction
Augmented reality (Augmented Reality, it is called for short AR) technology, it is a kind of by new technique integrated to real world information and virtual world information " seamless ", it is the entity information (visual information originally being difficult to experience in the certain time spatial dimension of real world, sound, taste, sense of touch etc.), by science and technology such as computers, superposition again after analog simulation, virtual information is applied to real world, by the perception of human sensory institute, thus reaches the sensory experience of exceeding reality.Real environment and dummy object have been added to same picture in real time or space exists simultaneously.
Augmented reality contains multimedia, three-dimensional modeling, real-time video show and control, Multi-sensor Fusion, real-time tracking and new technique and the means such as registration, scene fusion.Augmented reality provides in the ordinary course of things, is different from the appreciable information of the mankind.
FPAR (Fast Positioning Augmented Reality), FPAR are to integrate quickly location and the AR card of quick identification information.
2. card pattern is introduced
As shown in Figure 1, the method for designing of the FPAR card of the present embodiment is crossed the grain pattern shown in Fig. 1 and is carried out video identification, grain pattern is made up of 5 groups of concentric circulars/concentric polygon lines, and each group of concentric circular/concentric polygon is referred to as an anchor point, and 5 groups of concentric circular/concentric polygon referred to as identify grain pattern.The lines shape of often organizing anchor point is identical (can essentially be different, but in order to design convenient and more preferable recognition effect, make identical), wherein 4 groups of concentric circular/concentric polygon are arranged in the corner of rectangular card, 5th group of concentric circular/concentric polygon is scheduled on the middle position on the long limit of card, aligns with the center of circle of adjacent two grain patterns in its center of circle.(position of the most all concentric circular/concentric polygon is not necessarily intended to so arrange, it is also possible to be arranged on elsewhere, and after unique requirement simply they arrangements, total figure can not be similar to centrosymmetric grain pattern.During design, grain pattern can not be similar to centrosymmetry, can correctly identify the anglec of rotation when being for follow-up three-dimensional identification.) the grain pattern arrangement design reasons of above 5 groups of concentric circular/concentric polygon is easy for calculating below.
2.1. the printing of grain pattern/display requirement
In design, the number of plies of concentric circular/concentric polygon is infinite, but actually the least due to inner ring, it is possible to determine the number of plies according to printer or the printing of display and display precision, and the inner ring striped beyond printing precision or display capabilities scope can omit.In order to accurately identify, general requirement at least wants the black stricture of vagina of twice.The scaling at each circle edge isAs shown in Figure 2.
3, image-recognizing method
3.1, video input
The video collected from photographic head or other video source, first resolves into frame sequence.According to the operational capability situation of machine, can all identify with each frame or extract a frame every several frames and be identified.
3.2, RGB/YUV pretreatment
The frame extracted is probably RGB or yuv format.
When rgb format, method for quickly identifying is only to take redness or green channel, and rest channels is abandoned.Owing to card is black and white, so the passage so obtained still can restore the gray scale grain pattern of card.Owing to the compression ratio of blue channel can be heightened by the compressed format of a lot of videos, thus the definition of blue channel is the highest, so not using blue channel；Recommending red green weighting synthesis passage, this passage precision is higher, and operand is moderate.High-precision standard handovers is 0.299*R+0.587*G+0.114*B；Rapid translating can take RGB average.
When extracting yuv format, as long as then Y passage, abandon U, V passage.
By the step for process, result is that coloured image is become gray level image.
3.3, binary conversion treatment
The straightforward procedure of analysis quick to gray level image is become black white image exactly.General binary conversion treatment i.e. sets a threshold values, and brightness exceedes threshold values and is 1, is then 0 less than threshold values.In actual use, owing to identify grain pattern under different photoenvironments, so threshold values is difficult to predefine.Owing to uneven illumination is even, often can not find a suitable threshold values and be correctly separated grain pattern.So the present invention proposes a kind of dynamic, flexible method, pattern is carried out two-value separation.
This method needs to realize (the Blur filter will be discussed in 3.4) by the Blur filter.
By two copies of image copy, it is designated as copy 1 and copy 2.
Copy 1 carries out the Gaussian Blur (σ≤3px) of extremely trace, to eliminate the noise in video.If (efficiency to be improved, this step can be omitted).
Copy 2 carries out the Gaussian Blur that radius is bigger.Radius is 1/30 to 1/10 (empirical value calculates according to image length-width ratio 4:3) of image diagonal.Blur radius and the ability of camera lens, use environment, and even FPAR card prints material etc. relation, can set according to being actually needed.
Setting up a blank dot matrix image, size is consistent with front two copies, is designated as M, place brighter than copy 2 for correspondence position copy 1 is set to 1, is otherwise set to 0, then M is result.
The example of example 1: Fig. 3 shoots in the environment of ideal.Method mentioned above is used to process.As seen from Figure 3, in the result of binary conversion treatment, identify that grain pattern is very clear, reached preferable effect.
Example 2: Fig. 4 is the FPAR card example under ambisexuality state, card some under sunlight, another part under shade, use method mentioned above process.Binaryzation result from above is not difficult to find out, even if the FPAR card of artwork is negative and positive, but the result after binaryzation is the most highly desirable, is affected entirely without by negative and positive.It practice, what the parameter of each step determined that, identical with example 1 effect.The amount of Gaussian Blur, and the threshold value of final step binaryzation can predefine, and will not change because negative and positive environment changes.
Example 3: in order to show the advantage of this method, still by artwork in the same manner as above, uses traditional straightforward procedure to carry out binaryzation.I.e. set threshold values, when brightness is less than certain threshold values, become black, be otherwise white.In Fig. 5, the anchor point that a figure only one of which is complete, other anchor point is all blocked or partial occlusion；The anchor point part in the b figure upper right corner is blocked；C figure is then that the anchor point in the lower right corner be can't see.During identification, a figure is higher by threshold value setting, and b figure is by threshold value setting in centre, and c figure must be higher by threshold value setting.The experiment proved that, use traditional method no matter threshold value setting is become how many, be difficult to obtain more satisfied result.By discussed above, simple binarization method cannot adapt to these complex situations, and threshold value sets extremely difficult.Algorithm the most in this paper substantially has advantage.
Example 4: contrast is low, the extreme case that noise is many.By Fig. 6, the place of white, card after treatment occurs in that too much black speck originally；Originally the place of black, card after treatment has the color of white.And in any case adjustment threshold value all can not reach preferable effect.
Fig. 7 lists the result applying this algorithm process, and wherein, leftmost figure is equivalent to the copy 1 in 3.3, has carried out slight Gaussian Blur noise reduction, and other steps are completely the same with the first two example.From Fig. 7 (d) it can be seen that identify that grain pattern, black circle is full, white circle is clean, identifies that the local periphery at grain pattern place is more conducive to the identification in later stage almost without noise, this most relatively conventional threshold values algorithm.
Example 5: the situation of photographic fog.
FPAR card Ridge tracing has shake advantage.The picture of Fig. 8 intercepts from a dynamic video, and now FPAR card has bigger rocking, and therefore creates fuzzy picture.After carrying out binary conversion treatment by context of methods, the black circle of outmost turns is the most visible, and the grain pattern therefore obtained still can be supported to identify.
And if with the Ridge tracing of traditional method identification FPAR card, not being used in combination the method for the present invention when carrying out binary conversion treatment, then the binaryzation result that cannot obtain.As it is shown in figure 9, from left to right threshold value is from high to low successively.It is clear in five width figures five the anchor point energy Correct Analysis not having a width figure.It can thus be appreciated that the binary conversion treatment of FPAR Ridge tracing and the present invention needs to be used in combination.If not using FPAR Ridge tracing, only use the binary processing method of the present invention, the recognition effect that also cannot obtain.
3.4, the Blur filter
3.4.1, basic skills
Following table lists the pluses and minuses of some conventional image processing methods.
3.4.2 the quick improvement optimized for mobile device
For the mobile device of current main flow, its operational capability is typically difficult to support method 1 and 2.Can consider to use the method 3 of simple coarse, and the preferable method of effect 4.Owing to method 4 is containing more floating-point operation, it is unfavorable for the quick calculating of mobile device, it is possible to make improvements to adapt to mobile device.
According to article " IIR Gaussian Blur Filter Implementation usingAdvanced Vector Extensions " and " Recursive Gaussian Derivative Filters " in description, it is possible to use the positive double iir filter of inverted order, wherein forward is P0=p0, other Pn=w0*pn+w1*Pn-1+w2*Pn-2, similarly, reverse form is Qlen-1=Plen-1, remaining Qn=w0*Pn+w1*Qn+1+w2*Qn+2.Wherein the part of subscript out of bounds uses the mode of clamper to process, and i.e. takes the legal subscript of its arest neighbors.Use iir filter that Gaussian Blur is emulated with such form.Wherein P, p, Q, and w are floating-point.Therefore often calculate a pixel, at least need to carry out 6 floating-point multiplications.Assume that the image resolution that photographic head obtains is 640*480, capture 25 frame per second, then need to carry out floating-point multiplication 640*480*6*25=46080000 time each second.This operand is undoubtedly a great expense incurred for mobile device.
The present invention proposes a kind of method, tables look-up with shaping, by the way of trading space for time, avoids floating-point operation and multiplying, obtains and the result of former floating-point operation method approximation.
First, w0,w1,w2These three floating point values, is determined by the radius of Gaussian Blur, once radius determines, these three value is can be precalculated.And P and p is the numerical value between 0-255, wherein P is floating-point, and p is integer.Therefore naturally can consider the result of product of all of P and p Yu w is carried out pre-stored, make a table, with use for future reference, it is therefore an objective to avoid floating-point multiplication.
Owing to the numerical value between single-precision floating point 0-255 is too many, will list and all be worth and prestore, memory headroom takies the hugest.So the way that present aspect uses a kind of approximation lists a less precomputation multiplication table.
Current main-stream processor all uses IEEE754 floating-point format, is characterized in that a bit sign is followed by some exponents, is finally mantissa.For 32 single-precision floating point numerical value, blocking only affect its precision by latter 16 of mantissa, logarithm value impact is the least.Therefore note P ' is the numerical value after P blocks 16, end, data P after blocking ' the most remaining 16, have 65536 kinds of possible values, may values and w by these 65536 kinds0, w1, and w2It is multiplied respectively, and product stored result, the memory headroom size taken is little, and this provides possibility for quick computation of table lookup.Approximate formula is P*w ≈ P ' * W=W [P '], p*w=W ' [p '], and wherein W and W ' is the precalculated result of product of storage of array.
Calculate in view of IIR and comprise floating-point addition operation, owing to floating add is more time-consuming than Integral additive operation.The present invention uses addition of integer to replace floating-point addition operation.Notice that this step is the improvement to the preceding paragraph, formula with above will have nuance.As a example by IIR positive sequence direction, owing to comprising fractional part in calculating, therefore employing integer calculations and calculate to simulate fractional fixed point, carry out integer calculations with 16 shorts, wherein integer and fractional part respectively account for 8.P* ω * 256=(p ＜ ＜ 8) * ω ≈ [p* ω * 256]=Ω ' [p]；P*ω≈[p*ω]≈Ω[p′].The most actually computing formula of the IIR of positive sequence is Pn=ω0*(pn<<8)+ω1*Pn-1+ω2*Pn-2, PnFor integer, result of calculation is about 256 times of former formula.During calculating, available Ω table short-cut multiplication computing, upper son becomes Pn=Ω ' [pn]+Ω[Pn-1]+Ω[Pn-2]。
The scale-up problem of positive numerical sequence can not process, and after inverted order has calculated, directly use moves to right 8 and can repair.The table that inverted order uses with positive sequence is consistent, simply computing formula slightly difference.For Qn=ω0*Pn+ω 1*Qn+1+ω2*Qn+2.Multiplication herein again may be by looking into Ω table and solves, and expression formula is Qn=Ω [Pn]+Ω[Qn-1]+Ω[Qn-2].Then Rn=Qn> > 8 it is required.
So far, all floating-point operations are all replaced with fixed-point calculation, and all multiplication all become table look at peek.The operation efficiency of mobile device is improved significantly and improves.
* further, it is possible to use the mode reducing identification layer resolution makes the operand of following three kinds of calculation step reduce:
1) color conversion；2) Gaussian Blur；3) color in.
Thus significantly improve the speed of overall operational.Such as artwork is 640*480, but change GTG the step for become 320*240, then subsequent step 1), 2), 3) operand reduce to original 1/4.When the result generating virtual reality carries out pinup picture synthesis, still by the resolution of 640*480.So order output image sharpness is maintained, and differing only in this and producing error is the pinup picture site error within 1 pixel, and general user's naked eyes are difficult to find out the difference in quality of image.
3.5, judging may anchor point
3.5.1, fill
Check each pixel of M, if current pixel is 0 or 1, then with numerical value 2, carry out 4 connections or 8 connections are filled.And add up this Breadth Maximum filled and height, and also have the area filled, wherein area is exactly pixel quantity specifically.If length-width ratio is (this is empirical value, specifically can adjust) between 1:4 to 4:1, and filling rate existsNear, then rectangular outer frame position and the size of this being filled scope recorded in list L.
3.5.2, region lambda limiting process is filled
When image is very big, the area of filling may be very big, and the figure of filling may be the most complicated, for tradition filling algorithm, the most on the mobile apparatus, it may occur that stack overflow or the bigger situation of committed memory.In order to solve this problem, can limit filling region, as set left and right up-and-down boundary.Even from (x, y) starts to fill, set maximum and fill region u, then left margin is x-u, and right margin is x+u, and coboundary is y-u, and lower boundary is y+u.So can limit total filling area less than 4u2, thus alleviate internal memory burden.The method is referred to as filling region lambda limiting process.
3.5.3 the meaning of padding
Content in list L, owing to having following special features, it is possible to be elected to be possible locating point position.
First, the anchor point of FPAR card is circular, and each circle of the inside is all a circle, thus the most undistorted without spin under the conditions of, the FPAR card anchor point photographed also is positive round, and now using completion method to obtain the length-width ratio of housing should be 1:1.
But due to FPAR card may slant setting, so in the picture photographed, anchor point is not necessarily circular, and likely distort and present similar ellipse, so length-width ratio is not necessarily 1:1, and certain distortion leeway should be reserved, empirically, this proportion threshold value should arrange more relaxed.Such as can limit be between Aspect Ratio is at 1:4 to 4:1 legal.
Further, since the design of anchor point is graftal, so either which laminated striation sample, its filling rate is all consistent.Should be close to 0.436.
Two essential conditions being to judge anchor point above, therefore all legal anchor points all can occur in L, but also there are some flase drop points, need to utilize method below to reject.
3.5.4, fill frame and select example
Figure 10-11 gives an example filling frame choosing.The region of two kinds of oblique lines actually represents the same numerical value, is intended merely to show clear and obtain difference.The region that this step that what " " hatched example areas represented is just has been inserted, and " // " hatched example areas represent be before the region once filled of step.Figure 10 only lists part steps.
3.5.5. concentric frame
Owing to anchor point is made up of concentric circular/concentric polygon grain pattern, so will necessarily be capped repeatedly above coloring in step when.If so certain position is real anchor point in picture, then internal concentric/the concentric polygon at this place, position is inherently repeated several times in L and is chosen, and sees Figure 12.
By utilizing this rule, comparing the item in L, if some frame is independent, or other any frame of getting along well is concentric, then may be considered flase drop.Most flase drop point can be excluded with this.By the center x of each group of concentric frame, y-coordinate is recorded, and is designated as table H.If picture exists discernible complete grain pattern, then H is including at least the information of 5 anchor points in grain pattern, but sometimes due to flase drop, is also possible to there is the information more than 5 suspicious anchor points in H, now need further to be judged, get rid of the anchor point of flase drop.
Additionally, the set of the frame of each possible anchor point is respectively arranged with feature in composition H, when anchor point is more than 5, can be by the comprehensive concentricity examining frame group, length-width ratio similarity and tuple may to give a mark by anchor point for each, record in the lump in H, for being used behind.If H counting less than 5, then judge that there is not legal FPAR card in picture identifies grain pattern.
3.5.6. the situation of possible flase drop
During all identification described above, only give the sufficient condition of anchor point identification, not necessarily condition.So actually there may be the anchor point situation of various flase drop.As shown in figure 13, may detect more than 5 anchor points.But through the most various screenings and checking, typically it still is able to obtain correct result.
Figure 13 lists some patterns that may cause flase drop.Figure 14 lists the non-circular pattern that may result in flase drop of three kinds of structures；
Owing to the pattern similar with anchor point is more, so in actual applications, the flase drop of anchor point is unavoidable.Although the erroneous judgement situation of anchor point can be improved by increasing criterion, but it is complicated, so that separately seek method to make to detect process exception.In actual applications, multiple flase drop point may occur simultaneously, but with FPAR card of the present invention, the arrangement mode of flase drop point identifies that the probability that in grain pattern, the arrangement mode of anchor point is consistent is the most small.Thus the localization method that is given of the present invention can guarantee that and reaches gratifying effect.
If deliberately found, the pattern of mahjong bobbin may cause flase drop, as shown in figure 15.
3.5.7. from anchor point to FPAR card grain pattern identification
In 3.5.4, no matter H counts more than 5 or is just 5 points, all there is a problem in that the anchor point in the FPAR card identification grain pattern that each anchor point in H is the most corresponding？Now need to use permutation and combination method, enumerate all of correspondence and substitute into computing formula (the 4th introduction later), the most reasonable to test which kind of corresponded manner.
3.5.8. the impact on recognition effect of the region lambda limiting process is filled
When using filling region lambda limiting process, the maximum outer ring of possible some anchor points of wrong identification.But owing to grain pattern has scaling feature [2.2.1], so when can correctly identify internal layer some laminated striations sample, correct recognition result can be produced equally.
In figure 16, owing to employing a less filling border as region limits.The centre of form filling border overlaps with the position filling seed.After using region limits method, finding out (seeing Figure 17) from result figure, the black region of pattern outmost turns has been coated with 5 kinds of colors, has i.e. been divided into five and has selected frame (step 1,2, shown in 3,4,5), and the white portion of secondary outer ring has been also partitioned into three kinds of colors, i.e. divide into three and select frame (step 6, shown in 7,8).Totally 8 are selected frame to above 5+3 is all invalid, but enclose the most each layer from the second layer is black, and it is complete correctly to enclose choosing, thus the most effective.Internal layer black and white adds up and is of five storeys altogether, has been enough to be used in identifying.
Showing the frame selection condition not using region limits, totally 7 layers of ring in Figure 18, can analyze complete 7 laminated striation samples, 7 layers may be incorporated for identifying.
For anchor point identification, for avoiding flase drop, typically will at least 2 layers.Although the most results of the number of plies are the most accurate, but often reaching the precision of pixel scale after the number of plies to three or four layers, having had little significance up adding, as long as so can recognize that after restricted area that about three layers just can reach acceptable effect.
If the filling border of region limits method arranges narrow, it is possible to create impact have:
1) frame selects that the number of plies is very few causes recognition failures
2) anti-float ability weakens, and justs think as the situation of example 5 in 3.3, if owing to the reason of region limits and can not identify outermost black concentric circular/concentric polygon, will result directly in whole grain pattern recognition failures.
4. 2 d-3 d transformation approach
4.1. 2 d-3 d changes each label definition
The present invention this assumes that camera lens is undistorted, as camera lens exists pincushion distortion or barrel distortion, then should first correct, reapply following method.
FPAR is stuck in the three dimensions of reality, puts with any anglec of rotation, and its five anchor points are designated as ABCDE, as shown in figure 19 respectively.FPAR card camera-shot arrives, and obtains the process of bidimensional image, can be regarded as each point on FPAR card and is connected with O point, then meets at imaging picture.In imaging picture, each anchor point of FPAR card is designated as abcde.
The most each anchor point judges
Due to each point on FPAR card, a point in which point corresponding imaging picture, which point corresponding b point, which point corresponding c, d, e point etc., it is all unknown, moreover H may comprise flase drop point, see Figure 20.So to sort and Weeding, on FPAR card, the position relationship between a, b, c, d, and e point has a following evident characteristic:
(1) polygon abcd should be convex polygon；
(2) ∠ bec should be close to 180 °；
(3) distance of e to b and c should be closer to.
Specifically judge that step is as follows:
(1) convex polygon is first judged
Judge that convex polygon method is as follows:
C. get rid of line segment and have intersection option, see Figure 21.
D. product jack per line is judged
A, b, c, d, and e are regarded as three-dimensional point, wherein z-axis=0, then calculates Value, these values whether with axle jack per line, and can might as well all be set to positive sign.
(2) Comprehensive Evaluation
Owing to each anchor point is by image recognition, so above-mentioned 2,3 regular judgment criteria should be flexible.Thus during operation, be not directly to evaluate whether to meet the requirements, but it is how many for evaluating the laminating degree required with rule.So should 3 rules of summary, determine rational standards of grading.Mark is the highest, represents and more meets the requirements.
(3) Optimal Decision-making
When operational capability is permitted, appointing a little in H, can be chosen 5 and carry out fully intermeshing, substitute into above rule to verify, if the candidate items in H is too many, fully intermeshing is an astronomical figure, in order to alleviate operand, higher front n the point of H mid score (n >=5) can be taken and carry out substituting into test.
Through calculated above, it is possible to select in H 5 points meeted the requirements most, and obtain the corresponding relation of they and abcde each point.
4.3. the 3D approach vector of FPAR card is obtained
It is now to by the anti-normal vector pushing away FPAR card of imaging pictureRender serving follow-up virtual reality.To this end, cross O point to make boost line l1//AD//BC；l2//AB//DC；Note OAB normal vector isOCD normal vector isOBC normal vector isOAD normal vector isThe normal vector of note FPAR card isl1Direction vector beDirection vector be
Then have:
Finally give
Wherein × represent vector multiplication cross.
IfDirection is far off the beam, and the FPAR card such as calculated is almost vertical with view plane, then need to reject this result, and return 4.2, other permutation and combination of gravity treatment.
4.4.Z the many solutions of axle not impact analysis
As shown in figure 22, pushed away FPAR card by two-dimensional imaging picture is counter, because of z-axis Location-Unknown, it is possible that there is solve more.Little FPAR card the most nearby seems actually do not have difference with big FPAR card at a distance.Imaging picture is in front difference on rear simply scaling.If it practice, simply uniocular vision, this distance is unmeasured.But the difference on these differences simply scaling, does not impacts calculating below.So can set viewpoint the most simply to the distance of imaging picture is n, viewpoint is m to the distance at FPAR card center, in order to follow-up calculating.
4.5. proving correctness
Distance according to 4.4 is it is assumed that and 4.3Can be by the three-dimensional position of the anti-ABCDE point released in three dimensions on FPAR card in the position of the abcde point in the imaging picture of plane.Calculation is as follows:
By a French, by usingDistance with O to FPAR card, it can be deduced that the equation of FPAR card place plane Σ is lxx+lyy+lz(z-m)=0；
The intersection point of ray Oa, Ob, Oc, Od, and Oe and Σ is the three-dimensional position at ABCDE place.
As a example by A point, remember ta=lzm/(lxax+lyay+lzaz)；Then the coordinate of A is (ta*ax, ta*ay, ta*az).In like manner can try to achieve the coordinate of BCDE each point.
Being apparent from from above formula, each point coordinates all has product m, so might as well take m is a non-zero definite value, as taken 2, then takes az=n=1.So coordinate a little can calculate concrete numerical value, in order to subsequent calculations.
By the three-dimensional position analysis to ABCDE point, it is judged that result correctness.If numerical value is far off the beam, such as ∠ ABC differs greatly with space right-angle, or E point position differs greatly with BC point midway, or the length-width ratio of the rectangle of the anchor point composition on four angles is the most remote with FPAR card design phase difference, then judge that this result is incorrect, need to reject this result, and return other possible permutation and combination of 4.2 gravity treatments.
Step 4.5,4.3,4.2 is one group of relation verified mutually, needs to push away the most mutually just to obtain best result.
Furthermore it is also possible to add the verification method of some auxiliary.Such as, if the Aspect Ratio difference of each frame is relatively big in a doubtful anchor point, then consider that this anchor point is probably flase drop point.In like manner, if identifying that in grain pattern, in one group of anchor point, the mean aspect ratio of each ring is inconsistent, then it is assumed that this identifies that grain pattern may be incorrect.It practice, what the normal vector of FPAR card that the expectation length-width ratio of anchor point can be by above calculating calculated, if value of calculation differs greatly with measured value, can also determine that this identification is incorrect.
Be identified according to integrated approach all of the above, clear at image and do not have can not anti-interference (such as picture occurs in that two FPAR cards simultaneously) time, the recognition result drawn can reach correct close to 100%.In examples given below, background is newspaper, and on newspaper, grain pattern is complicated, and interference is big, but does not the most affect recognition result.
If same picture has multiple FPAR cards, then will have a lot of anchor point and be detected simultaneously, but which five anchor point will belong to same group of very difficult resolution on earth.Finally perceive due to computer is not a pictures, but one group of suspicious anchor point, it is possible that the suspicious anchor point in certain figure may be as shown in figure 23.
Distribution situation as Figure 23, probably which naked eyes all cannot tell is flase drop point, and which is true point, which anchor point one FPAR card grain pattern of composition.Now, the grain pattern of FPAR card can be made improvements, such as, concentric circular be changed and be C-shaped.A kind of improvement project for reference is as shown in figure 24: may be just like the effect of Figure 25 after then multiple FPAR cards are put together.
The anchor point of this grain pattern can be identified by the algorithm of the present invention equally, simply they the most information of ratio, it is simply that the direction attribute of FPAR card.Doing example by a leftmost card, we can identify the opening direction of C-shaped mouth, as shown in figure 26.
Except opening direction, we can also calculate the credibility of this opening direction by numerical value such as the definition of detection opening, angles.Opening direction and the double information of credibility, we the direction of arrow the most indicated by an arrow can represent that most probable opening direction, arrow size represent credibility, see Figure 27.
Upper figure has had directional information, for previous figure of only locating point position, has been easy to find out which is real anchor point at a glance, and which anchor point can form one and consist of FPAR grain pattern.Can judge according to following principle:
Many FPAR card identification method with direction:
1) paying the utmost attention to opening direction point with a high credibility is anchor point.
2) if the direction of two anchor points is close and close with the direction of their line, then they may belong to same FPAR card, and is belonging respectively to B, E point.Wherein B point points to E point.
3) if from certain anchor point, subsequent point is found by the direction of each point, 5 anchor points can be found the most continuously, and the direction credibility of last anchor point is significantly lower than front 4 points, then 5 anchor points of this continuous print probably belong to the ABECD anchor point of same FPAR card.
4) nearby principle, if there being two other anchor points in certain anchor point direction, then pays the utmost attention to belong to same FPAR card with its nearer.
Time actually used, due to the distortion of camera, the bending of FPAR card, accuracy of identification is inadequate, and camera lens is out of focus or shake causes the reasons such as fuzzy pictures, and the vector result drawn there may be error, so direction said before is pointed to, it is all to judge the degree of approximation rather than must be directed at the poorlyest.So being also required to carry out certain permutation and combination when measuring, it is the most rational to test which kind of coupling.But by 4 rule above, we can carry out significantly beta pruning to enumeration process, greatly reduce test number (TN).
5. render virtual reality
Actual virtual reality renders, and can be the solid adding three-dimensional in picture, it is also possible to only stick pinup picture.
Figure 28 is that the FPAR card multi-angle imaging of a 3-D view realizes example.
It is now to distort the picture preset, is then filled with in picture.In order to increase practicality and sense of reality, the photo inserted needs to hide all anchor points, no matter and FPAR card how to put, photo seem will FPAR card relatively static, it appears that just as really having a photo plaster on FPAR card.In order to accomplish this point, committed step is intended to the three-dimensional position by FPAR card, extrapolates its every bit position in imaging picture, sees Figure 28.Method in this calculating process and 4.5 is a pair inverse operation.
Being attached to by a sheet of planar picture virtually on three-dimensional FPAR card, calculate after picture is attached on FPAR card, the color of each corresponding point should show that position on imaging plane.
Practical operation needs from being rendered to image plane, searches corresponding point position on FPAR card, and this calculating is more complicated, and border is difficult to make, and operand is the biggest.So needing to use other succinct methods to replace.Simplest method is the method using drawTriangles, not exclusively calculate on FPAR card color a little, but do a sparse matrix grid, then by this mesh mapping to imaging plane.The conversion of the position triangular fragments in the hole of grid is simulated.The scheme of a kind of triangle section is as shown in figure 29.
DrawTrianges or its different name equivalent method directly supporting by each platform, such as HTML5, OpenGL, FLASH ActionScript3, cocos2D-x etc., also by the support of video card rapid computations.It it is general and conventional calculation method.Specifically refer to following network address (explanation of FLASH ActionScript3 version) http://help.adobe.com/zh_CN/as3/dev/WS84753F1C-5ABE-40b1-A2E4-0 7D7349976C4.html.
In order to increase sense of reality, under FPAR card is in different light environment, the tone of pinup picture should be adjusted, to allow it coincide with the tone of FPAR card.Having the empty position much not having anchor point on FPAR card, the color that can take these positions carries out white balance to pinup picture, and brightness etc. adjusts, to allow its color and luster laminating actual environment.
Finally supplement and render example below:
Example 6: prepare three and clap the vertical photos that to obtain, copies the FPAR card clapping vertical photograph shape fabricating, as shown in picture on Figure 30 desktop for one.
Being real-time rendering effect on Figure 30 right plate computer, in figure, FPAR card can be mixed the spurious with the genuine, screen is seen just as if four clap and vertical to obtain photo.Figure 31 is for rendering sectional drawing effect.
Figure 32 represents the expanded function of FPAR card, not only photo is rendered in picture in figure, has also rendered hand-written " Guangzhou 2016 ", and has not identified grain pattern accordingly on " Guangzhou 2016 " position.It can be seen that the region rendered can extend.Additionally it is noted that in real-time rendering example, there is a footmark in the photo upper right corner, the position of footmark even renders outside photo frame, and this also embodies the autgmentability of its rendering capability.
Above-described embodiment is the present invention preferably embodiment; but embodiments of the present invention are also not restricted to the described embodiments; the change made under other any spirit without departing from the present invention and principle, modify, substitute, combine, simplify; all should be the substitute mode of equivalence, within being included in protection scope of the present invention.
Claims (10)
1. the method for designing of a FPAR card, it is characterised in that described FPAR card is provided with many groups with one heart
Circle/concentric polygon lines, each group of concentric circular/concentric polygon is referred to as an anchor point, organize more concentric circular/
Concentric polygon composition identifies grain pattern, and described that to organize total figure after concentric circular/concentric polygon arrangement be center is more
Symmetrical.
2. the method for designing of follow-up FPAR card described in claim 1, it is characterised in that described FPAR
Card is provided with 5 groups of concentric circulars/concentric polygon lines, and the lines shape often organizing anchor point is identical, wherein 4 groups
Concentric circular/concentric polygon is arranged in the corner of rectangular card, and the 5th group of concentric circular/concentric polygon is scheduled on card
The middle position on long limit or be on the asymmetric position being capable of identify that FPAR card；Often group group concentric circular/with
The scaling at heart polygon two adjacent rings edge is
3. the quick recognition positioning method of a FPAR card, it is characterised in that comprise the steps:
S1, video input, first resolve into frame sequence by the video source file collected；
S2, RGB/YUV pretreatment, it is RGB or yuv format that the frame extracted is probably, and works as extraction
Arrive for rgb format time, method for quickly identifying is only to take redness or green channel, and rest channels is lost
Abandon, when extract for yuv format time, as long as then Y passage, abandon U, V passage；
S3, binary conversion treatment, carry out two-value separation to image；
S4, employing the Blur filter method carry out image recognition processing；Use the double iir filter of positive inverted order, its
Middle forward is P0=p0, other Pn=w0*pn+w1*Pn-1+w2*Pn-2, similarly, reverse form is Qlen-1=Plen-1,
Remaining Qn=w0*Pn+w1*Qn+1+w2*Qn+2, wherein the part of subscript out of bounds uses the mode of clamper to process,
I.e. take the legal subscript of its arest neighbors, use iir filter that Gaussian Blur is emulated with such form；
S5, judge possible anchor point,
S6, the 3D approach obtaining FPAR card are vectorial, by the anti-normal vector pushing away FPAR card of imaging picture, with
Serve follow-up virtual reality to render；
S7, by comparison and drift correction, determine the three-dimensional position of each anchor point, so that it is determined that whole FPAR
It is stuck in three-dimensional position；
S8, render virtual reality, virtually a sheet of planar picture is attached on three-dimensional FPAR card, calculates figure
After sheet is attached on FPAR card, the color of each corresponding point should show that position on imaging plane.
The quick recognition positioning method of FPAR card the most according to claim 3, it is characterised in that institute
State in step S3, realize binary conversion treatment by the Blur filter, itself particularly as follows:
By two copies of image copy, it is designated as copy 1 and copy 2；
If there being more apparent noise in video, copy 1 carries out the Gaussian Blur of extremely trace, wherein σ≤3px；
Copy 2 carries out the Gaussian Blur that radius is bigger, and radius is 1/30 to the 1/10 of image diagonal；
Setting up a blank dot matrix image, size is consistent with front two copies, is designated as M, by correspondence position copy 1
The place brighter than copy 2 is set to 1, is otherwise set to 0, then M is result.
The quick recognition positioning method of FPAR card the most according to claim 3, it is characterised in that institute
State in step S4, carry out image procossing method particularly includes:
S41, iir filter by simulation Gaussian Blur determine w0, w1, w2 these three floating point values, and P
And the numerical value that p is between 0-255, wherein P is floating-point, and p is integer；Therefore by part P and p and w
Result of product carry out pre-stored, make a table, with use for future reference；
S42, the way of a kind of approximation is used to list a less precomputation multiplication table；
Approximate formula is P*w ≈ P ' * W=W [P '], and p*w=W ' [p], wherein W and W ' is storage of array
Precalculated result of product, P ' is the result after single precision floating datum P blocks 16 mantissa bits.
The quick recognition positioning method of FPAR card the most according to claim 3, it is characterised in that institute
State in step S4, use addition of integer to replace floating-point addition operation, in IIR positive sequence direction, owing to calculating
In comprise fractional part, therefore employ integer calculations and calculate to simulate fractional fixed point, carry out with 16 shorts
Integer calculations, wherein integer and fractional part respectively account for 8, The most actually computing formula of the IIR of positive sequence is Pn=ω0*(pn＜ ＜ 8)+ω1*Pn-1+ω2*Pn-2, PnFor integer, result of calculation is about 256 times of former formula, during calculating, and available Ω table
Short-cut multiplication computing, above formula becomes Pn=Ω ' [pn]+Ω[Pn-1]+Ω[Pn-2]；
The scale-up problem of positive numerical sequence can not process, and directly uses and move to right 8 after inverted order has calculated
Can repair, the table that inverted order uses with positive sequence is consistent, simply computing formula slightly difference, for Qn=ω0*Pn+ω1*Qn+1+ω2*Qn+2, multiplication herein again may be by looking into Ω table and solves, and expression formula is
Qn=Ω [Pn]+Ω[Qn-1]+Ω[Qn-2], then Rn=Qn＞ ＞ 8 is required.
The quick recognition positioning method of FPAR card the most according to claim 3, it is characterised in that step
Rapid S5 particularly as follows:
S5.1, each pixel of inspection M, if current pixel is 0 or 1, then with numerical value 2, carry out 4
Connection or 8 connections are filled, and add up this Breadth Maximum filled and height, also have the area filled；
S5.2, to fill region limit；Even from (x, y) start fill, set maximum fill region u,
Then left margin is x-u, and right margin is x+u, and coboundary is y-u, and lower boundary is y+u；So can limit total
Filling area less than 4u2, thus alleviate internal memory burden；
S5.3, selected possible locating point position, can limit and be conjunction between Aspect Ratio is at 1:4 to 4:1
Method, the design additionally, due to anchor point is graftal, so either which laminated striation sample, its filling rate is all
Consistent, and filling rate is near 0.436；The point only meeting above-mentioned requirements just counts form L；
S5.4, eliminating flase drop point, compare to the item in form, if some frame is independent, or
Person's other any frame of getting along well is concentric, then may be considered flase drop；The concentric frame that selects of each group in form L is carried out
Merging, record its average central and mark stores to table H, its mid score needs consideration to relate to this group
The concentricity of each frame, length-width ratio concordance, frame overlap number and each select frame filling rate, more meet the requirements, point
Number is the highest, otherwise the lowest；
S5.5, from anchor point to FPAR card grain pattern identification, the method using permutation and combination, it is all of fixed to enumerate
Site also substitutes into computing formula, the most reasonable to test which kind of corresponded manner.
The quick recognition positioning method of FPAR card the most according to claim 7, it is characterised in that step
In rapid S5.4, for anchor point identification, for avoiding flase drop, typically will at least 2 layers.
The quick recognition positioning method of FPAR card the most according to claim 3, it is characterised in that step
Rapid S5 particularly as follows: each anchor point judgement particularly as follows:
(1) convex polygon is first judged
Judge that convex polygon method is as follows:
A. get rid of line segment and have intersection option, it is judged that product jack per line；
A, b, c, d, and e are regarded as three-dimensional point, wherein z-axis=0, then calculates Value, these values whether with axle jack per line, and can might as well all be set to positive sign；
(2) Comprehensive Evaluation；
Owing to each anchor point is by image recognition, during operation, it not directly to evaluate whether to meet the requirements, but
It is how many for evaluating the laminating degree required with rule；
(3) Optimal Decision-making；
When operational capability is permitted, appointing a little in H, can be chosen 5 and carry out fully intermeshing, substitute into above
Rule verify, if the candidate items in H is too many, fully intermeshing is an astronomical figure, in order to alleviate
Operand, front n the point that can take H mid score higher carries out fully intermeshing substitution test.
The quick recognition positioning method of FPAR card the most according to claim 3, it is characterised in that step
Rapid S7 particularly as follows: the method that uses drawTriangles, not exclusively calculate on FPAR card face a little
Color, but do a sparse matrix grid, then by the hole of grid in this mesh mapping to imaging plane
The conversion of position triangular fragments be simulated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610137618.1A CN105894068B (en) | 2016-03-10 | 2016-03-10 | FPAR card design and rapid identification and positioning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610137618.1A CN105894068B (en) | 2016-03-10 | 2016-03-10 | FPAR card design and rapid identification and positioning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105894068A true CN105894068A (en) | 2016-08-24 |
| CN105894068B CN105894068B (en) | 2020-07-28 |
Family
ID=57014486
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610137618.1A Active CN105894068B (en) | 2016-03-10 | 2016-03-10 | FPAR card design and rapid identification and positioning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105894068B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108955983A (en) * | 2018-07-25 | 2018-12-07 | 湖南大学 | Cable tension test method based on the drag-line vibration shape and photogrammetric technology |
| CN110674806A (en) * | 2019-10-31 | 2020-01-10 | 威创集团股份有限公司 | Automatic generation method and device of AR identification card |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040028258A1 (en) * | 2002-08-09 | 2004-02-12 | Leonid Naimark | Fiducial detection system |
| JP2010061409A (en) * | 2008-09-03 | 2010-03-18 | Kyodo Printing Co Ltd | Image processing program and image processing system |
| CN101866496A (en) * | 2010-06-04 | 2010-10-20 | 西安电子科技大学 | Augmented reality method based on concentric ring pattern group |
| CN104636779A (en) * | 2013-11-11 | 2015-05-20 | 覃政 | Annular code identifier recognition system |
-
2016
- 2016-03-10 CN CN201610137618.1A patent/CN105894068B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040028258A1 (en) * | 2002-08-09 | 2004-02-12 | Leonid Naimark | Fiducial detection system |
| JP2010061409A (en) * | 2008-09-03 | 2010-03-18 | Kyodo Printing Co Ltd | Image processing program and image processing system |
| CN101866496A (en) * | 2010-06-04 | 2010-10-20 | 西安电子科技大学 | Augmented reality method based on concentric ring pattern group |
| CN104636779A (en) * | 2013-11-11 | 2015-05-20 | 覃政 | Annular code identifier recognition system |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108955983A (en) * | 2018-07-25 | 2018-12-07 | 湖南大学 | Cable tension test method based on the drag-line vibration shape and photogrammetric technology |
| CN110674806A (en) * | 2019-10-31 | 2020-01-10 | 威创集团股份有限公司 | Automatic generation method and device of AR identification card |
| CN110674806B (en) * | 2019-10-31 | 2022-08-16 | 威创集团股份有限公司 | Automatic generation method and device of AR identification card |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105894068B (en) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hu et al. | Revisiting single image depth estimation: Toward higher resolution maps with accurate object boundaries | |
| US10540576B1 (en) | Panoramic camera systems | |
| Farid | Photo forensics | |
| AU2017246470B2 (en) | Generating intermediate views using optical flow | |
| CN106683068B (en) | Three-dimensional digital image acquisition method | |
| Sen et al. | Robust patch-based hdr reconstruction of dynamic scenes. | |
| JP2022173399A (en) | Image processing apparatus, and image processing method | |
| CN106780546B (en) | The personal identification method of motion blur encoded point based on convolutional neural networks | |
| CN105335950A (en) | Image processing method and image processing apparatus | |
| CN105303615A (en) | Combination method of two-dimensional stitching and three-dimensional surface reconstruction of image | |
| CN113205502A (en) | Insulator defect detection method and system based on deep learning | |
| CN109360144B (en) | Image real-time correction improvement method based on mobile phone platform | |
| Liu et al. | Exemplar-based image inpainting with multi-resolution information and the graph cut technique | |
| EP3309750B1 (en) | Image processing apparatus and image processing method | |
| Xu et al. | Scalable image-based indoor scene rendering with reflections | |
| CN116342519A (en) | Image processing method based on machine learning | |
| Yang et al. | Doing more with Moiré pattern detection in digital photos | |
| CN105894068A (en) | FPAR card design method and rapid identification and positioning method of FPAR card | |
| CN117011493B (en) | Three-dimensional face reconstruction method, device and equipment based on symbol distance function representation | |
| CN109785429A (en) | A kind of method and apparatus of three-dimensional reconstruction | |
| CN114066715A (en) | Image style migration method and device, electronic equipment and storage medium | |
| CN106548184A (en) | A kind of method and apparatus of adjustment illumination balance | |
| CN113052311B (en) | Feature extraction network with layer jump structure and method for generating features and descriptors | |
| Tola | DAISY: A Fast Descriptor for Dense Wide Baseline Stereo and Multiview Reconstruction. | |
| Han et al. | Guided filtering based data fusion for light field depth estimation with L0 gradient minimization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |