CN109785367A - Exterior point filtering method and device in threedimensional model tracking - Google Patents
Exterior point filtering method and device in threedimensional model tracking Download PDFInfo
- Publication number
- CN109785367A CN109785367A CN201910052458.4A CN201910052458A CN109785367A CN 109785367 A CN109785367 A CN 109785367A CN 201910052458 A CN201910052458 A CN 201910052458A CN 109785367 A CN109785367 A CN 109785367A
- Authority
- CN
- China
- Prior art keywords
- color
- point
- model
- tracking
- profile
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention provides exterior point filtering method and device in a kind of tracking of threedimensional model, and wherein method includes: to extract profile in current frame image or extract characteristic point；It obtains based on configuration sampling point or based on the color model of characteristic point；Next frame image is read in, the initial position of profile or characteristic point in the next frame is predicted according to motion model；Optimize profile or the corresponding characteristic point of search in the next frame according to initial position；Based on color model, the energy value of configuration sampling point or characteristic point color probability distribution is calculated, and exterior point is filtered out according to the energy value of calculating；Next frame posture is solved, and next frame is replaced with into present frame.The present invention is able to solve threedimensional model and tracks bloom in actual scene, complex environment and the problems such as block, to improve the success rate and robustness of three-dimensional tracking.
Description
Technical field
Embodiment of the present invention is related to the exterior point side of filtering out in computer vision field more particularly to a kind of tracking of threedimensional model
Method and device.
Background technique
Tracking technique early stage is mainly used in plane tracking, specifically in a certain amount of video sequence, for spy
Fixed target (such as vehicle, people, billboard etc.) is tracked, and marks the position of target in video in the video sequence.
Target following has a wide range of applications in computer vision field: such as video monitoring, vehicle flowrate monitoring, unmanned, face
Identification, augmented reality (augmented reality, AR) etc..Such as in recognition of face, need to grab in the video sequence in real time
The position where face is taken, the relevant feature of face could be extracted and identified；In the field AR, need in real time in video sequence
The middle position marked that grabs out renders etc..
Above-mentioned tracking generally directed to be specific one piece of pixel region or planar object in image.With computer
It calculates the promotion of power and the demand in market, tracking technique is gradually excessive to three-dimension object.In the tracking of three-dimension object, in addition to needing
Position of the three-dimension object in video pictures is indicated in real time, it is also necessary to calculate the size of three-dimension object, direction etc. is detailed
Posture information.Especially in the field AR, if posture information is not accurate enough, the dummy object rendered will very great Cheng
User experience is influenced on degree.
The U.S. Patent application US20140369557 of the prior art 1 is disclosed a kind of is for the tracking based on feature
System and method, wherein what trace mode was taken is NCC prediction+plane IC iteration+signature tracking (feature tracking).
There are in the case where template, the approximate region where coarse search determines target is carried out first with NCC, then from flat in image
Face extracted region is singly answered (homography) out and is changed in the way of inverse combination (Inverse Compositional)
In generation, further obtains more accurate position, finally on this basis, using feature tracking, Feature Points Matching
Mode determines final accurate position.In the system that the patent defines, the output of position when containing series of computation exception,
For example when feature tracking failure, export the position etc. of Inverse Compositional iteration.
The Chinese patent application CN 106845435 of the prior art 2 discloses a kind of expansion based on detection tracing algorithm in kind
Increase real border Implementation Technology, what is mainly taken is to preset three dimensional practicality template and posture, is tracked.Specifically with
During track, a regional frame is divided to tracing area, and characteristic point is extracted in region.If characteristic point quantity meets threshold
Value demand then carries out the tracking of characteristic point.Its major way is to search when present frame extracts ORB characteristic point in next frame searching
The corresponding characteristic point of rope, to estimate the pose of the threedimensional model of next frame.If characteristic point quantity is inadequate, extract in region
Edge, utilize edge contour optimization estimation threedimensional model posture (pose).
But in implementing the present invention, it may, inventor has found the prior art, at least there are the following problems:
For the prior art 1, the mode of plane iteration is taken, it is necessary to which in tracking sequence, existing characteristics point is abundant flat
Otherwise face region just can be can cause failure.In addition, the tracking of three-dimension object does not often meet the assumptions of plane, asking
Also problem is easy to appear in solution preocess.
For the prior art 2, in the scheme for taking characteristic point, since three-dimension object is when visual angle converts, originally
Visible angle point or face may become invisible, easily cause tracking failure.It is switched to when object texture is less based on edge
When tracking, the problems such as due to various illumination or complex background, edge probably converges to local optimum, solves and makes mistake
Pose causes subsequent tracking to fail.
In addition, in the tracking of threedimensional model, since partial 3-D model lacks texture, and it is reflective etc. there are bloom, it is difficult
To generate the tracking that fixed template carries out characteristic point, to calculate target in real time often by area tracking or Edge track
Posture, facilitate and carry out subsequent processing.But since the Fatal defects such as object module itself and ambient lighting are readily incorporated
Exterior point；It is widely present and blocks in practical applications, traditional algorithm is relatively easy to introduce exterior point.Above-mentioned phenomenon is all easy to cause calculation
Posture devious or cause to lose at method solution.
It should be noted that the above description of the technical background be intended merely to it is convenient to technical solution of the present invention carry out it is clear,
Complete explanation, and facilitate the understanding of those skilled in the art and illustrate.Cannot merely because these schemes of the invention
Background technology part is expounded and thinks that above-mentioned technical proposal is known to those skilled in the art.
Summary of the invention
In view of the above-mentioned problems, embodiment of the present invention is designed to provide the exterior point side of filtering out in a kind of tracking of threedimensional model
Method and device are able to solve threedimensional model and track bloom in actual scene, complex environment and the problems such as block, to improve
The success rate and robustness of three-dimensional tracking.
To achieve the above object, embodiment of the present invention provides exterior point filtering method in a kind of tracking of threedimensional model, comprising:
Present frame extracts profile；Obtain the color model based on configuration sampling point；Next frame image is read in, according to the pre- measuring wheel of motion model
Wide initial position in the next frame, and profile is optimized according to initial position in the next frame；Based on color model, calculate
The energy value of configuration sampling point color probability distribution, and exterior point is filtered out according to the energy value of calculating；Next frame posture is solved, and will
Next frame replaces with present frame.
Specifically, in the scheme based on profile, according to the posture of present frame, the profile of three-dimension object is projected into two dimension
On image, wherein assuming that image profile occurs in the corresponding position of profile in the algorithm, thus to solve profile.
The color model be each point in the edge of projected outline color probability distribution model or the overall situation color it is general
Rate distributed model.
The distribution of color in the preset range centered on configuration sampling point is counted, each point in configuration sampling region is counted
Color value, color value is quantified, the probability that each color occurs in quantization rear profile sampling area is calculated, is formed each
The color probability distribution model of point；The color probability distribution mould of each point of contour edge is recalculated after the tracking of each frame
Type, or with the color probability distribution model of the preset rate update each point of contour edge.Alternatively, statistics all samplings of profile
Distribution of color in the preset range of point, quantifies color value, and each color goes out in configuration sampling region after calculating quantization
Existing probability forms the color probability distribution model of the profile overall situation；Recalculate that profile is all to be adopted after the tracking of each frame
The probability distribution of sampling point forms global color probability Distribution Model；Alternatively, updating global color probability distribution with preset rate
Model.
Next frame image is read in, predicts that the configuration sampling point that previous frame extracts in the next frame may according to motion model
The initial position of appearance searches for edge in the next frame according to initial position to optimize edge.
To achieve the above object, embodiment of the present invention also provides exterior point filtering method in a kind of tracking of threedimensional model, packet
Include: present frame extracts characteristic point；Obtain the color model based on characteristic point；Next frame image is read in, is predicted according to motion model
The position of characteristic point, and search for corresponding characteristic point in the next frame according to initial position；Based on color model, calculate special
The energy value of sign point probability distribution, and exterior point is filtered out according to the energy value of calculating；Next frame posture is solved, and next frame is replaced
For present frame.
Specifically, in the scheme based on characteristic point, the posture tracked according to 3-D image, which determines, to be needed in image
The region of middle calculating, extracts angle point in region and corresponding description of calculating generates characteristic point.
The color model be each characteristic point color probability distribution model or all characteristic points global color it is general
Rate distributed model.
The distribution of color in the preset range centered on each characteristic point is counted, color value is quantified, calculation amount
The probability that each color occurs after change, forms the color probability distribution model of each characteristic point；The weight after tracking of each frame
The color probability distribution model of each characteristic point is newly calculated, or updates the color probability point of each characteristic point with preset rate
Cloth model.Alternatively, counting the distribution of color in all characteristic point preset ranges, color value is quantified, is calculated every after quantifying
The probability that kind color occurs, forms the global color probability Distribution Model of characteristic point；It is recalculated after the tracking of each frame
The probability distribution of all characteristic points forms the global color probability Distribution Model of characteristic point；Alternatively, being updated with preset rate special
Levy the global color probability Distribution Model of point.
Next frame image is read in, is likely to occur in the next frame according to the characteristic point that motion model predicts that previous frame extracts
Initial position；Search for corresponding characteristic point in the next frame according to initial position；Alternatively, extracting feature near prime area
It puts and description is utilized to compare and search for corresponding characteristic point.
Embodiment of the present invention also provides exterior point filtering device in a kind of tracking of threedimensional model, in the threedimensional model tracking
Exterior point filtering device, which is performed, realizes exterior point filtering method in foregoing threedimensional model tracking.
Embodiment of the present invention also provides exterior point filtering device in a kind of tracking of threedimensional model, including memory and processing
Device, in which: memory is used for store code and document；Processor is for executing the code and document stored in the memory
To realize exterior point filtering method in foregoing threedimensional model tracking.
Therefore exterior point filtering method and device in the threedimensional model tracking of embodiment of the present invention offer, by hair
The many experiments of bright people are studied, and calculate current tracking point in real time by the history color model of statistics tracking point or this trace regions
Energy, propose the formula of completely new calculating energy value E, wherein E is smaller, and the correct probability of sampled point is bigger so that according to
Energy value E, and pass through setting energy threshold or other adaptive threshold process modes, it will be able to easy and quasi-ly by mistake
Profile or characteristic point to being filtered out, avoid the profile of error tracking or characteristic point from impacting whole tracking effect, from
And improve threedimensional model is tracked under each scene stability and robustness.It especially, can be in the scheme based on profile
Bigger weight is assigned for the sampled point closer from profile in energy function, so result will be made more reliable.
Detailed description of the invention
It, below will be to embodiment in order to illustrate more clearly of embodiment of the present invention or technical solution in the prior art
Or attached drawing needed to be used in the description of the prior art is simply introduced one by one, it should be apparent that, the accompanying drawings in the following description is
Some embodiments of the present invention, for those of ordinary skill in the art, without creative efforts, also
Other drawings may be obtained according to these drawings without any creative labor.
Fig. 1 is the flow diagram for the 3-D image method for tracing based on characteristic point that embodiment of the present invention provides；
Fig. 2 is the flow diagram for the 3-D image method for tracing based on profile that embodiment of the present invention provides；
Fig. 3 is the flow diagram of exterior point filtering method during the threedimensional model that embodiment of the present invention provides is tracked；
Fig. 4 is the schematic diagram that 3-D image tracking is carried out based on profile that embodiment of the present invention provides；
Fig. 5 is the flow diagram of exterior point filtering method during the threedimensional model that embodiment of the present invention provides is tracked；
Fig. 6 is the structural schematic diagram of exterior point filtering device during the threedimensional model that embodiment of the present invention provides is tracked.
Specific embodiment
To keep the purposes, technical schemes and advantages of embodiment of the present invention clearer, implement below in conjunction with the present invention
The technical solution in embodiment of the present invention is clearly and completely described in attached drawing in mode, it is clear that described reality
The mode of applying is some embodiments of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ability
Domain those of ordinary skill every other embodiment obtained without creative efforts, belongs to the present invention
The range of protection.
Embodiment of the present invention provides exterior point filtering method in a kind of tracking of threedimensional model, wherein in 3-D image tracking
It can be carried out based on characteristic point and based on profile.
As shown in Figure 1, a kind of method of 3-D image tracking can be based on profile, generally comprise:
101, the projection of present frame three-dimensional object profile on the image is calculated；
102, next frame image is read in；
103, the position of profile transformation is predicted according to motion model；
104, next frame image border is extracted, with contour fitting Optimization Solution；And
105, next frame posture is solved, and next frame is replaced with into present frame.
As shown in Fig. 2, another method of 3-D image tracking can be based on characteristic point, generally comprise:
201, present frame extracts characteristic point and stores；
202, next frame image is read in；
203, according to motion model predicted characteristics point position；
204, search characteristics point；And
205, next frame posture is solved, and next frame is replaced with into present frame.
Exterior point filtering method in the threedimensional model tracking that embodiment of the present invention provides, based on characteristic point and based on profile
3-D image tracing scheme on the basis of, and introduce new module based on color model, exterior point can be effective filtered out, thus
The problems such as solving bloom of the threedimensional model tracking in actual scene, complex environment and blocking, further promotes three-dimension object and chases after
The robustness of track.
As shown in figure 3, exterior point filtering method in threedimensional model tracking should be realized based on color model, following step is specifically included
It is rapid:
Step 301, present frame extracts profile or extracts characteristic point.
It is carried out in the scheme of 3-D image tracking based on profile is extracted:
According to the posture of present frame, the profile of three-dimension object is projected on two dimensional image, wherein algorithm assumes that image exists
The corresponding position of profile edge should occur to solve profile, such as by the position at global optimization edge or can pass through method
Line search finds the high position of gradient as edge sample point, so that profile is solved, at this to the specific method for solving profile
And with no restriction.
In the scheme for carrying out 3-D image tracking based on characteristic point:
The posture (pose) tracked according to 3-D image determines the region for needing to calculate in the picture.It is mentioned in region
It takes angle point and calculates and generate corresponding Feature Descriptor.Wherein extracting angle point can be FAST (features from
Accelerated segment test) angle point, or Harris angle point or similar Robust Algorithm of Image Corner Extraction, at this to mentioning
Take the specific method of angle point and with no restriction.
For the angle point extracted, calculates corresponding description and generate its characteristic point.The mode for generating Feature Descriptor can
Think that simple extract patch (patch) generates NCC (Normalized Cross Correlation)/SAD (Sum of
Absolute Differences)/SSD (Sum of Squared Differences) etc. matching template；It may be ORB
(Oriented FAST and Rotated BRIEF) description or SIFT (Scale Invariant Feature
Transform)/SURF (Speed Up Robust Feature) etc. description son, at this to description son and be not specifically limited.
Step 2, it obtains based on configuration sampling point or based on the color model of characteristic point.
In this step, counting statistics or update are based on configuration sampling point or based on the color model of characteristic point.
Color model is the color probability distribution model of each point in the edge of profile or the color probability distribution mould of the overall situation
Type, color here can refer to the color model in each channel RGB of script computer normal image format storage, may also mean that
RGB color model transforms to the color model of HSV/HSI etc..RGB color model: rgb color mode is a kind of face of industry
Colour standard, be by red (R), green (G), the variation of blue (B) three Color Channels and their mutual superpositions come
To miscellaneous color, RGB is the color for representing three channels of red, green, blue, this standard almost includes mankind's view
The all colours that power can perceive are current with most wide one of color system.Hsv color model: HSV (Hue,
Saturation, Value) it is a kind of color space created by A.R.Smith in 1978 according to the intuitive nature of color,
Claim hexagonal pyramid model (Hexcone Model).The parameter of color is respectively in this model: tone (H), and saturation degree (S) is bright
It spends (V).HSV can be converted mutually with RGB.
Since there may be subtle color differences for the pixel at the same position of interframe, in order to consider efficiency and robustness,
Color channel values can generally be quantified, for example each channel value of script RGB is 0~255, i.e., a total of 65535 kinds of colors,
It can be quantified as 512 or 256 kind of color, HSV/HIS is also similarly.Quantification manner can be adopted in different Color Channels
Different quantized values is taken, such as in HSV, for considering for color discrimination, the channel H can be quantified thinner.
Color probability distribution model can refer to the probability distribution of single feature point or edge sample point neighboring area color,
It can also refer to that all the points neighboring area is gathered together the global color model to be formed.For the mode of characteristic point, due to feature
The uncertainty of point distributing position, often takes independent color model.For the mode at edge, since sampled point is gathered in side
Near edge, its whole color model is often counted, and divide color model to prospect (interior of articles) model and background model point
It does not store.
Probability distribution refers to that in the color model of current region, the probability that this color occurs, wherein all colours go out
Existing probability summation is 1.Probability distribution can be what each frame generated in real time, i.e., recalculates feature after each frame tracking
The color probability distribution model for putting perhaps edge sample point can also form binding with this feature point or edge sample point and close
System updates this probability distribution after generating initial color probability distribution model with given pace, and such benefit is can be with
The error for preventing picture suddenly change from introducing.
By taking RGB color model as an example, it is assumed that there are two types of color, a kind of 60% region is (255,0,0) for target area, is remained
Lower 40% field color is (0,255,0)；A total of n pixel in target area.In original RGB model, Mei Getong
A total of 256 kinds of values in road, are quantified as m value for each channel.Then Color Channel amounts to m3Kind distribution.Enable j=256/m.Then divide
Cloth is ((0~j-1), 0~(j-1), 0~(j-1)), ((1~2j-1), 0~(j-1), 0~(j-1)) ````````````
((256-j)~255, (256-j)~255, (256-j)~255)；Wherein, distribution of color ((256-j)~255,0~(j-1),
0~(j-1)) value be 0.6n, (0~(j-1) 0~(j-1), (256-j)~255) value be 0.4n.It is respectively 0.6 He after normalization
0.4, residual value 0.These values reflect the probability that the color within the scope of corresponding color occurs in target area.
In the scheme for carrying out 3-D image tracking based on profile, the color model based on configuration sampling point is obtained, it is specific to wrap
It includes:
The distribution of color in the preset range centered on configuration sampling point is counted, each point in configuration sampling region is counted
Color value, color value is quantified, the probability that each color occurs in quantization rear profile sampling area is calculated, is formed each
The color probability distribution model of point；The color probability distribution mould of each point of contour edge is recalculated after the tracking of each frame
Type, or with the color probability distribution model of the preset rate update each point of contour edge；
Alternatively,
The distribution of color in the preset range of all sampled points of profile is counted, color value is quantified, after calculating quantization
The probability that each color occurs in configuration sampling region forms the color probability distribution model of the profile overall situation；It is tracked in each frame
After recalculate all sampled points of profile probability distribution formed global color probability Distribution Model；Alternatively, with preset
Rate updates global color probability Distribution Model.
In the scheme for carrying out 3-D image tracking based on characteristic point, the color model based on characteristic point is obtained, is specifically included:
The distribution of color in the preset range centered on each characteristic point is counted, color value is quantified, calculation amount
The probability that each color occurs after change, forms the color probability distribution model of each characteristic point；The weight after tracking of each frame
The color probability distribution model of each characteristic point is newly calculated, or updates the color probability point of each characteristic point with preset rate
Cloth model；
Alternatively,
The distribution of color in all characteristic point preset ranges is counted, color value is quantified, calculates every kind of face after quantization
The probability that color occurs, forms the global color probability Distribution Model of characteristic point；It is recalculated after the tracking of each frame all
The probability distribution of characteristic point forms the global color probability Distribution Model of characteristic point；Alternatively, updating characteristic point with preset rate
Global color probability Distribution Model.
Step 3, next frame image is read in, profile or predicted characteristics point position are predicted according to motion model.
In this step, read in next frame image, according to motion model predict the edge sample point that extracts of previous frame or
The initial position that person's characteristic point is likely to occur in the next frame, to facilitate algorithm search optimization profile or characteristic point.
Motion model can be the acceleration of history posture, or the initial position that simple coarse search provides, Huo Zheyou
The initial displacements of the offers such as equipment gyroscope and rotation etc. to motion model and are not specifically limited at this.
Step 4, optimize profile or the corresponding characteristic point of search in the next frame according to initial position.
In the scheme for carrying out 3-D image tracking based on profile:
Profile can be determined simply by along normal search gradient direction, Edge Distance field also can be generated, by excellent
The penalty function for changing distance field obtains correct outline position, to the specific method for solving profile and with no restriction at this.
By taking Fig. 4 as an example, when tracking threedimensional model, wherein inside frame (sampled point) is previous frame object external outline foundation
Previous frame posture projects the profile of present frame, and outside frame (sampled point) is the profile of the present frame obtained by searching method
Figure.402 images are the middle graph of the front and back scape segmentation in treatment process, and 403 images are the edge image of present frame, 404 images
For the distance field image in distance field optimization.
In the scheme for carrying out 3-D image tracking based on characteristic point:
It can be the search of the template matchings such as simple SAD/SSD/NNC, ORB/ can also be extracted near prime area
The characteristic points such as SIFT/SURF utilize description to compare and determine corresponding characteristic point, simultaneously to the specific method of determining characteristic point at this
With no restriction.
Step 5, based on color model, the energy value of configuration sampling point or characteristic point color probability distribution is calculated, and
Exterior point is filtered out according to the energy value of calculating.
In the scheme for carrying out 3-D image tracking based on profile:
Due to the influence of complex background or bloom etc., false contouring may be generated near true profile.False contouring
With gradient, the algorithm for directly searching class may allow it to judge by accident.And these false contourings can generate interference on distance field,
It allows in algorithmic statement to false contouring, therefore is also required to filter out the sampled point near false contouring.
By taking Fig. 5 as an example, it is shown that the attitude misalignment caused by false edges " deception ", wherein due to multiple in 501 images
Miscellaneous background, the lines that can be generated in false contouring, such as 503 images inside wire frame near true profile are exactly false contouring, meeting
Cause to optimize mistake, it is therefore desirable to filter out these false contourings.
In the energy function definition based on profile, it is as follows to define energy function E:
Wherein, Ic(x) pixel value under present frame at x coordinate is indicated；
Ω is the sampling area of configuration sampling point；
Pr(Ic(x)) posterior probability of the current color in the color model of global prospect is indicated；
Pb(Ic(x)) posterior probability of the current color in the color model of global background is indicated；
φ (x) is Signed Distance Field, and when x is in prospect, the positive distance of return returns negative when x is located at background
To distance.
HeIt (x) is smooth heavisde function, the bigger power of the closer point of distance sample in imparting sampling area
Weight, the smooth function that other can also be taken linear certainly, and be not specifically limited.
Under the limitation of the global color model of statistics, the marginal point being currently found is located at true energy function E in formula (1)
The probability at positive edge (and contour of object).The point of low probability is screened out by the way that threshold value is arranged, it is ensured that the remaining sampling searched
Perhaps optimization point is respectively positioned on real edge rather than on vacation edge caused by mixed and disorderly background or bloom etc. point, to mention
The subsequent robustness for solving posture is risen.
Further, since the distribution of configuration sampling point is similar all on profile, global statistics can be taken to carry out interior point outer
The screening of point.Furthermore, it is possible to assign bigger weight for the sampled point closer from profile in energy function, will so make to tie
Fruit is more reliable.
In the scheme for carrying out 3-D image tracking based on characteristic point:
Since view transformation characteristic point disappears, or due to blocking, bloom etc. causes Feature Points Matching mistake etc., mistake
Matching meeting is so that the posture finally solved generates deviation.
In the energy function definition based on characteristic point, it is as follows to define energy function E:
Wherein,Indicate region near this feature point；
X is pixel coordinate；
Ic(x) pixel value under present frame at x coordinate is indicated.
Energy function E in formula (2) intuitively measures color samples and history distribution of color needs one near characteristic point
It causes, in the case where good threshold is arranged in advance, can effectively filter out exterior point.
In the above-mentioned scheme based on profile or based on characteristic point progress 3-D image tracking, and P (Φ | Ω) bigger, table
Show that the point is bigger for the probability of actual profile point or characteristic point.Generally for the convenience of subsequent optimization and calculating, E is chosen as most
Energy expression afterwards, i.e., E is smaller, and the correct probability of sampled point is bigger.In general, it blocks if it exists or situations such as bloom
When, local color model can not be met with global color model, it will or a bigger E, if having converged to mistake
Edge, color model is also inconsistent with historical information, can also obtain a biggish E.Global color model or part
History color model etc. can choose different measures according to practical application scene.Then can be arranged according to calculated E simple
Threshold filtering, the E point for being greater than certain threshold value is considered as exterior point and filtered out, adaptive threshold process side can also be passed through
Formula, such as mean value add fixed numbers, or the fixed point etc. for deleting maximum 10% energy, are not particularly limited herein.
Step 6, next frame posture is solved, and next frame is replaced with into present frame.
In this step, after exterior point filters out, guaranteeing that remaining point is all the characteristic point correctly searched or side
Edge sampled point solves the posture of three-dimension object under present frame using modes such as PNP or Gaussian weighting marks, into next frame
Tracking.Next frame posture is solved using identical calculation method, and next frame is replaced with into present frame, circular treatment.
The step of various methods divide above, be intended merely to describe it is clear, when realization can be merged into a step or
Certain steps are split, multiple steps are decomposed into, as long as including identical logical relation, all in the protection scope of this patent
It is interior；To adding inessential modification in algorithm or in process or introducing inessential design, but its algorithm is not changed
Core design with process is all in the protection scope of the patent.
The embodiment of the invention also provides exterior point filtering device in a kind of tracking of threedimensional model, in the threedimensional model tracking
Exterior point filtering device, which is performed, realizes exterior point filtering method in foregoing threedimensional model tracking.
As shown in fig. 6, embodiment of the present invention also provides exterior point filtering device in a kind of tracking of threedimensional model, including storage
Device and processor, in which:
Memory 601 is used for store code and document；
Processor 602, for executing the code stored in the memory and document to realize foregoing three
Exterior point filtering method in dimension module tracking.
In the above-mentioned threedimensional model tracking being related in exterior point filtering device particular technique details and threedimensional model tracking
Exterior point filtering method is similar, therefore no longer specifically repeats.
It will be understood by those skilled in the art that implementing the method for the above embodiments is that can pass through
Program is completed to instruct relevant hardware, which is stored in a storage medium, including some instructions are used so that one
A equipment (can be single-chip microcontroller, chip etc.) or processor (processor) execute each embodiment the method for the application
All or part of the steps.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only
Memory), random access memory (RAM, Random Access Memory), magnetic or disk etc. are various can store journey
The medium of sequence code.
Therefore exterior point filtering method and device in threedimensional model tracking provided by the invention, by counting tracking point
Or the history color model of this trace regions calculates the energy of current tracking point in real time, by energy threshold by mistake edge or
Person's tracking point avoids the edge of error tracking or point from impacting whole tracking effect, is promoted under each scene to filtering out
The stability and robustness of threedimensional model tracking.
Each embodiment in this specification is described in a progressive manner, same and similar between each embodiment
Part may refer to each other, what each embodiment stressed is the difference with other embodiments.
Finally, it should be noted that being supplied to ability to the description of various embodiments of the invention above with the purpose described
Field technique personnel.It is that it is not intended to exhaustion or be not intended to and limit the invention to single disclosed embodiment.Institute as above
It states, various substitutions of the invention and variation will be apparent for above-mentioned technology one of ordinary skill in the art.Therefore,
Although having specifically discussed some alternative embodiments, other embodiment will be apparent or ability
Field technique personnel relatively easily obtain.The present invention is intended to include herein by discussion cross all substitutions of the invention, modification and
Variation, and fall in the other embodiment in the spirit and scope of above-mentioned application.
Claims (12)
1. exterior point filtering method in a kind of threedimensional model tracking characterized by comprising
Profile is extracted in current frame image；
Obtain the color model based on configuration sampling point；
Next frame image is read in, the initial position of profile in the next frame is predicted according to motion model, and exist according to initial position
Optimize profile in next frame；
Based on color model, the energy value of configuration sampling point color probability distribution is calculated, and filter according to the energy value of calculating
Except exterior point；
Next frame posture is solved, and next frame is replaced with into present frame.
2. exterior point filtering method in threedimensional model tracking according to claim 1, which is characterized in that described in present frame figure
Profile is extracted as in, is specifically included:
According to the posture of present frame, the profile of three-dimension object is projected on two dimensional image, wherein assuming that image exists in the algorithm
There is profile in the corresponding position of profile, thus to solve profile.
3. exterior point filtering method in threedimensional model according to claim 1 tracking, which is characterized in that the color model is
The color probability distribution model of each point in the edge of profile or the color probability distribution model of the overall situation；
The color model of the acquisition based on configuration sampling point, specifically includes:
The distribution of color in the preset range centered on configuration sampling point is counted, the face of each point in configuration sampling region is counted
Color value quantifies color value, calculates the probability that each color occurs in quantization rear profile sampling area, forms each point
Color probability distribution model；The color probability distribution model of each point of contour edge is recalculated after the tracking of each frame,
Or the color probability distribution model of each point of contour edge is updated with preset rate；
Alternatively,
The distribution of color in the preset range of all sampled points of profile is counted, color value is quantified, calculates after quantization every kind
The probability that color occurs in configuration sampling region forms the color probability distribution model of the profile overall situation；Terminate in the tracking of each frame
The probability distribution for recalculating all sampled points of profile afterwards forms global color probability Distribution Model；Alternatively, with preset rate
Update global color probability Distribution Model.
4. exterior point filtering method in threedimensional model tracking according to claim 1, which is characterized in that the reading next frame
Image predicts the initial position of profile in the next frame according to motion model, and optimizes wheel in the next frame according to initial position
Exterior feature specifically includes:
Next frame image is read in, is likely to occur in the next frame according to the configuration sampling point that motion model predicts that previous frame extracts
Initial position, search for edge in the next frame according to initial position to optimize profile.
5. exterior point filtering method in threedimensional model tracking according to claim 1, which is characterized in that described to be adopted based on profile
The energy value of the probability distribution of sampling point is calculated by energy function E, the calculation formula of the energy function E are as follows:
E=-log (P (Φ | Ω))
Wherein, Ic(x) pixel value under present frame at x coordinate is indicated；
Ω is the sampling area of configuration sampling point；
Pf(Ic(x)) posterior probability of the current color in the color model of global prospect is indicated；
Pb(Ic(x)) posterior probability of the current color in the color model of global background is indicated；
Φ (x) is Signed Distance Field, and when x is in prospect, the positive distance of return returns to negative sense when x is located at background
Distance；
The bigger weight of the closer point of distance sample in He (x) imparting sampling area.
6. exterior point filtering method in a kind of threedimensional model tracking characterized by comprising
Present frame extracts characteristic point；
Obtain the color model based on characteristic point；
Next frame image is read in, is searched in the next frame according to the position of motion model predicted characteristics point, and according to initial position
Corresponding characteristic point；
Based on color model, the energy value of characteristic point probability distribution is calculated, and exterior point is filtered out according to the energy value of calculating；
Next frame posture is solved, and next frame is replaced with into present frame.
7. exterior point filtering method in threedimensional model tracking according to claim 6, which is characterized in that the present frame extracts
Characteristic point specifically includes:
The posture tracked according to 3-D image determines the region for needing to calculate in the picture, and angle point is extracted in region and is counted
It calculates corresponding description and generates characteristic point.
8. exterior point filtering method in threedimensional model according to claim 6 tracking, which is characterized in that the color model is
The color probability distribution model of each characteristic point or the global color probability Distribution Model of all characteristic points；
The color model of the acquisition based on characteristic point, specifically includes:
The distribution of color in the preset range centered on each characteristic point is counted, color value is quantified, after calculating quantization
The probability that each color occurs, forms the color probability distribution model of each characteristic point；It is counted again after the tracking of each frame
The color probability distribution model of each characteristic point is calculated, or updates the color probability distribution mould of each characteristic point with preset rate
Type；
Alternatively,
The distribution of color in all characteristic point preset ranges is counted, color value is quantified, each color goes out after calculating quantization
Existing probability forms the global color probability Distribution Model of characteristic point；All features are recalculated after the tracking of each frame
The probability distribution of point forms the global color probability Distribution Model of characteristic point；Alternatively, updating the complete of characteristic point with preset rate
Office's color probability distribution model.
9. exterior point filtering method in threedimensional model tracking according to claim 6, which is characterized in that the reading next frame
Image according to the position of motion model predicted characteristics point, and searches for corresponding characteristic point according to initial position in the next frame, tool
Body includes:
Next frame image is read in, is likely to occur in the next frame according to the characteristic point that motion model prediction previous frame extracts first
Beginning position；
Search for corresponding characteristic point in the next frame according to initial position；Alternatively, extracting characteristic point and benefit near prime area
It is compared with description and searches for corresponding characteristic point.
10. exterior point filtering method in threedimensional model tracking according to claim 6, which is characterized in that described to be based on feature
The energy value of the probability distribution of point is calculated by energy function E, the calculation formula of the energy function E are as follows:
Wherein,Indicate region near characteristic point；
X is pixel coordinate；
Ic(x) pixel value under present frame at x coordinate is indicated.
11. exterior point filtering device in a kind of threedimensional model tracking, which is characterized in that exterior point filters out dress in the threedimensional model tracking
It sets to be performed and realizes exterior point filtering method step in the tracking of threedimensional model described in any one of claims 1 to 10.
12. exterior point filtering device in a kind of tracking of threedimensional model, which is characterized in that described device includes memory and processor,
Wherein:
The memory is used for store code and document；
The processor, for executing the code stored in the memory and document to realize that claims 1 to 10 is appointed
Method and step described in one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910052458.4A CN109785367B (en) | 2019-01-21 | 2019-01-21 | Method and device for filtering foreign points in three-dimensional model tracking |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910052458.4A CN109785367B (en) | 2019-01-21 | 2019-01-21 | Method and device for filtering foreign points in three-dimensional model tracking |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109785367A true CN109785367A (en) | 2019-05-21 |
| CN109785367B CN109785367B (en) | 2019-12-20 |
Family
ID=66500949
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910052458.4A Active CN109785367B (en) | 2019-01-21 | 2019-01-21 | Method and device for filtering foreign points in three-dimensional model tracking |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109785367B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112614161A (en) * | 2020-12-28 | 2021-04-06 | 之江实验室 | Three-dimensional object tracking method based on edge confidence |
| WO2023078052A1 (en) * | 2021-11-02 | 2023-05-11 | 中兴通讯股份有限公司 | Three-dimensional object detection method and apparatus, and computer-readable storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6192156B1 (en) * | 1998-04-03 | 2001-02-20 | Synapix, Inc. | Feature tracking using a dense feature array |
| CN1873656A (en) * | 2005-06-03 | 2006-12-06 | 中国科学院自动化研究所 | Detection method of natural target in robot vision navigation |
| US20130182894A1 (en) * | 2012-01-18 | 2013-07-18 | Samsung Electronics Co., Ltd. | Method and apparatus for camera tracking |
| CN103606186A (en) * | 2013-02-02 | 2014-02-26 | 浙江大学 | Virtual hair style modeling method of images and videos |
| US8761439B1 (en) * | 2011-08-24 | 2014-06-24 | Sri International | Method and apparatus for generating three-dimensional pose using monocular visual sensor and inertial measurement unit |
| CN106096577A (en) * | 2016-06-24 | 2016-11-09 | 安徽工业大学 | Target tracking system in a kind of photographic head distribution map and method for tracing |
| CN106845435A (en) * | 2017-02-10 | 2017-06-13 | 深圳前海大造科技有限公司 | A kind of augmented reality Implementation Technology based on material object detection tracing algorithm |
| CN108345881A (en) * | 2018-02-01 | 2018-07-31 | 福州大学 | A kind of document quality detection method based on computer vision |
| CN109102523A (en) * | 2018-07-13 | 2018-12-28 | 南京理工大学 | A kind of moving object detection and tracking |
-
2019
- 2019-01-21 CN CN201910052458.4A patent/CN109785367B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6192156B1 (en) * | 1998-04-03 | 2001-02-20 | Synapix, Inc. | Feature tracking using a dense feature array |
| CN1873656A (en) * | 2005-06-03 | 2006-12-06 | 中国科学院自动化研究所 | Detection method of natural target in robot vision navigation |
| US8761439B1 (en) * | 2011-08-24 | 2014-06-24 | Sri International | Method and apparatus for generating three-dimensional pose using monocular visual sensor and inertial measurement unit |
| US20130182894A1 (en) * | 2012-01-18 | 2013-07-18 | Samsung Electronics Co., Ltd. | Method and apparatus for camera tracking |
| CN103606186A (en) * | 2013-02-02 | 2014-02-26 | 浙江大学 | Virtual hair style modeling method of images and videos |
| CN106096577A (en) * | 2016-06-24 | 2016-11-09 | 安徽工业大学 | Target tracking system in a kind of photographic head distribution map and method for tracing |
| CN106845435A (en) * | 2017-02-10 | 2017-06-13 | 深圳前海大造科技有限公司 | A kind of augmented reality Implementation Technology based on material object detection tracing algorithm |
| CN108345881A (en) * | 2018-02-01 | 2018-07-31 | 福州大学 | A kind of document quality detection method based on computer vision |
| CN109102523A (en) * | 2018-07-13 | 2018-12-28 | 南京理工大学 | A kind of moving object detection and tracking |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112614161A (en) * | 2020-12-28 | 2021-04-06 | 之江实验室 | Three-dimensional object tracking method based on edge confidence |
| WO2023078052A1 (en) * | 2021-11-02 | 2023-05-11 | 中兴通讯股份有限公司 | Three-dimensional object detection method and apparatus, and computer-readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109785367B (en) | 2019-12-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107862698B (en) | Light field foreground segmentation method and device based on K mean cluster | |
| CN103400409B (en) | A kind of coverage 3D method for visualizing based on photographic head attitude Fast estimation | |
| Zhang et al. | Lighting and pose robust face sketch synthesis | |
| CN109344701A (en) | A kind of dynamic gesture identification method based on Kinect | |
| CN109544456A (en) | The panorama environment perception method merged based on two dimensional image and three dimensional point cloud | |
| CN106250895B (en) | A kind of remote sensing image region of interest area detecting method | |
| CN108875595A (en) | A kind of Driving Scene object detection method merged based on deep learning and multilayer feature | |
| CN108537239B (en) | Method for detecting image saliency target | |
| CN108921895B (en) | Sensor relative pose estimation method | |
| CN106599836A (en) | Multi-face tracking method and tracking system | |
| CN104850850A (en) | Binocular stereoscopic vision image feature extraction method combining shape and color | |
| CN110070580A (en) | Based on the matched SLAM fast relocation method of local key frame and image processing apparatus | |
| CN109410316A (en) | Method, tracking, relevant apparatus and the storage medium of the three-dimensional reconstruction of object | |
| CN111242864A (en) | Finger vein image restoration method based on Gabor texture constraint | |
| CN104123554A (en) | SIFT image characteristic extraction method based on MMTD | |
| CN107886471A (en) | A kind of unnecessary object minimizing technology of photo based on super-pixel Voting Model | |
| CN110807787B (en) | Skyline extraction method and system | |
| CN109785367A (en) | Exterior point filtering method and device in threedimensional model tracking | |
| CN106874848A (en) | A kind of pedestrian detection method and system | |
| CN109146952A (en) | Estimate the method, apparatus and computer readable storage medium of compartment void volume | |
| CN109242854A (en) | A kind of image significance detection method based on FLIC super-pixel segmentation | |
| CN105184802A (en) | Image processing method and device | |
| CN103680145B (en) | A kind of people's car automatic identifying method based on local image characteristics | |
| CN111161219B (en) | Robust monocular vision SLAM method suitable for shadow environment | |
| CN105184809A (en) | Moving object detection method and moving object detection device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |