CN109801316A - A kind of top fruit sprayer three-dimensional point cloud automation method for registering and reconstructing method - Google Patents
A kind of top fruit sprayer three-dimensional point cloud automation method for registering and reconstructing method Download PDFInfo
- Publication number
- CN109801316A CN109801316A CN201811558734.6A CN201811558734A CN109801316A CN 109801316 A CN109801316 A CN 109801316A CN 201811558734 A CN201811558734 A CN 201811558734A CN 109801316 A CN109801316 A CN 109801316A
- Authority
- CN
- China
- Prior art keywords
- registration
- point cloud
- point
- parameter
- cloud
- 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)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The embodiment of the present invention provides a kind of top fruit sprayer three-dimensional point cloud automation method for registering and reconstructing method, this method comprises: default registration parameter, is iterated optimization according to preset rules, utilizes the registration parameter progress point cloud registering of optimization；Registration parameter includes first registration parameter, purification match parameter and smart registration parameter；Preset rules are registration effect evaluation index, according to priority orders, including NARF key point Duplication, NARF key point dispersion ratio and NARF key point range error；To source point cloud and the target point cloud after first registration be iterated it is smart be registrated, when meeting default preferentially parameter, stop iteration, acquisition two panels point cloud essence registration result.The embodiment of the present invention optimizes registration parameter by preset registration effect evaluation index, the point cloud chart picture registration of automation can be achieved, difference cloud can not be suitable for the defect of registration by overcoming existing same registration parameter, substantially increase the precision of registration, full-automation is realized, registration efficiency is improved.
Description
Technical field
The present embodiments relate to technical field of image processing, and in particular to a kind of automation of top fruit sprayer three-dimensional point cloud is matched
Quasi- method and reconstructing method.
Background technique
Three Dimensional Reconfiguration is the key technology in computer vision technique, it using computer to real goal scene into
Digitized is reproduced, analyzes and is handled.The three-dimensional information of scene more meets the visual characteristic of the mankind than two-dimensional image information, really
The Three Dimensional Reconfiguration of target scene has obtained more and more concerns and application.
Three-dimensional modeling is carried out to target object based on three-dimensional laser scanning system and needs a large amount of man-machine interactively, and exists and obtains
Take information rate compared with the problems such as slow, data redundancy and equipment price are high.Three-dimensional modeling based on body-sensing camera is needed to target
Object is manually shot, because of the uncontrollability of equipment and shooting environmental, it is difficult to ensure that each amplitude point cloud image maintains like or closely
As regard the elevation angle, illumination condition and the depth of field, when being registrated the point cloud data that shooting obtains to carry out three-dimensional modeling, same
It is different larger with punctual low precision to carrying out to difference cloud under one method for registering and parameter, therefore cause difference cloud to registration
Parameter is different, needs manually adjustment trial repeatedly, and image registration efficiency is lower, precision cannot be guaranteed to be best.
Summary of the invention
To solve the problems, such as that the degree of automation is low in prior art point cloud registering, precision is not high, the embodiment of the present invention is provided
A kind of top fruit sprayer three-dimensional point cloud automation method for registering and reconstructing method.
In a first aspect, the embodiment of the present invention provides a kind of top fruit sprayer three-dimensional point cloud automation method for registering, this method packet
Include: default registration parameter is iterated according to preset rules, optimizes the registration parameter, utilizes the registration parameter of optimization
Carry out point cloud registering；The registration parameter includes first registration parameter, purification match parameter and smart registration parameter；Wherein, described first
Registration parameter is for determining the minimum interest value of NARF key point, space support region and for calculating NARF key point
The search field radius of FPFH characteristic value, the purification match parameter are for purifying and matching the corresponding pass of NARF key point feature
The range deviation threshold value of system, the essence registration parameter are maximum corresponding points distance and the range error variation for smart registration point cloud
Threshold value；The preset rules are registration effect evaluation index, and according to priority orders, the registration effect evaluation index includes
NARF key point Duplication, NARF key point dispersion ratio and NARF key point range error；The NARF key point Duplication is
Ensure the NARF key point number after matching relationship purifies with point quantity in enough；The NARF key point dispersion ratio is NARF
Dispersion after the purification of key point matching relationship between interior point；The NARF key point range error is right in preliminary registration result
Answer the root mean square of the Euclidean distance between NARF key point；Wherein, the interior point refers in just registration process according to just with quasi-mode
Calculated range error meets the NARF key point of preset condition after type matching；To the source point cloud and target obtained after first registration
Point cloud is iterated smart registration, when meeting default preferentially parameter, stops iteration, obtains two panels point cloud essence registration result.
Second aspect, the embodiment of the present invention provide a kind of top fruit sprayer three-dimensional point cloud Automated Refactoring, this method packet
It includes: three-dimensional to adjacent view top fruit sprayer using the automation method for registering of top fruit sprayer three-dimensional point cloud described in the embodiment of the present invention
Point cloud two-by-two in certain sequence converted all the points cloud by the intermediate conversion matrix by essence registration, preservation intermediate conversion matrix
Space coordinates to where last visual angle point cloud, obtain top fruit sprayer three-dimensional point cloud；The adjacent view point cloud, refers to
It is not less than 30% consecutive points cloud using the two panels point cloud Duplication obtained during Kinect acquisition fruit tree three dimensional point cloud；
The intermediate conversion matrix refers to the registration transition matrix from source point cloud spatial alternation to target point cloud space.
The embodiment of the present invention optimizes registration parameter, it can be achieved that the point cloud automated by preset registration effect evaluation index
Image registration, difference cloud can not be suitable for the defect of registration by overcoming same registration parameter in the prior art, be greatly improved
The precision of registration, and full-automation is realized, improve the registration efficiency of two panels point cloud.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is the present invention
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 is that top fruit sprayer three-dimensional point cloud provided in an embodiment of the present invention automates method for registering flow chart；
Fig. 2 is top fruit sprayer three-dimensional point cloud Automated Refactoring flow chart provided in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, the technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Fig. 1 is that top fruit sprayer three-dimensional point cloud provided in an embodiment of the present invention automates method for registering flow chart.Such as Fig. 1 institute
Show, which comprises
Step 101, default registration parameter, are iterated according to preset rules, optimize the registration parameter, utilize optimization
The registration parameter carries out point cloud registering；The registration parameter includes first registration parameter, purification match parameter and smart registration parameter；
Wherein, the just registration parameter is for determining the minimum interest value of NARF key point, space support region and for calculating
The search field radius of the FPFH characteristic value of NARF key point, the purification match parameter are for purifying and matching NARF key
The range deviation threshold value of point feature corresponding relationship, it is described essence registration parameter be for smart registration point cloud maximum corresponding points distance and
Range error change threshold；
The preset rules are registration effect evaluation index, according to priority orders, the registration effect evaluation index packet
Include NARF key point Duplication, NARF key point dispersion ratio and NARF key point range error；The NARF key point Duplication
To ensure the NARF key point number with point quantity interior enough after matching relationship purification；The NARF key point dispersion ratio is
Dispersion after the purification of NARF key point matching relationship between interior point；The NARF key point range error is preliminary registration result
The root mean square of Euclidean distance between middle corresponding NARF key point；Wherein, the interior point refers to just in registration process according to just matching
Calculated range error meets the NARF key point of preset condition after quasi- Model Matching；
Step 102, to the source point cloud obtained after first registration and target point cloud be iterated it is smart be registrated, preset preferentially meeting
When parameter, stop iteration, obtains two panels point cloud essence registration result.
Based on the source point cloud and target point cloud obtained after first registration, for each point in source point cloud, in target point cloud
Find the corresponding points composition point Cloud transform point set that nearest and distance is less than the maximum respective distances of smart registration point cloud；It utilizes
Singular value decomposition method calculates rotational transformation matrix and translation transformation matrix of the point Cloud transform point concentration between source to target, goes forward side by side
Row is from source point cloud to the transformation of target point cloud；Carry out calculating after iteration twice according to the above method smart registration point cloud range error it
Difference, the range error change threshold if more than smart registration point cloud then continue iteration, until the distance for meeting smart registration point cloud is missed
Until poor change threshold condition.
The embodiment of the present invention optimizes registration parameter, it can be achieved that the point cloud automated by preset registration effect evaluation index
Image initial registration, difference cloud can not be suitable for the defect of registration, significantly by overcoming same registration parameter in the prior art
The precision of registration is improved, and realizes full-automation, improves the registration efficiency of two panels point cloud.
Further, based on the above embodiment, the NARF key point dispersion ratio chooses the very poor of NARF key point y-coordinate
The sum of very poor with z coordinate indicates:
S=| ymax-ymin|+|zmax-zmin|
Wherein, s indicates the NARF key point dispersion ratio, ymaxAnd yminThe respectively maximum value of y-coordinate axis direction and most
Small value, zmaxAnd zminThe respectively maximum value and minimum value of z coordinate axis direction.
Affiliated NARF key point dispersion ratio is the dispersion after NARF key point matching relationship purifies between interior point；Due to fruit
Use Kinect sensor using vertical direction as axis rotary taking counterclockwise when tree three-dimensional point cloud acquisition, therefore the correspondence
NARF key point dispersion ratio chooses the sum of the very poor expression of the very poor and z coordinate of key point y-coordinate.
On the basis of the above embodiments, the embodiment of the present invention is closed by introducing NARF key point dispersion ratio to evaluate NARF
Dispersion degree between key point is avoided since NARF key point excessively concentrates the erroneous judgement caused to registration result；In after purification
Dispersion ratio is bigger between point, can more show a structure for cloud entirety, point cloud registering effect is better.
Further, based on the above embodiment, the NARF key point range error is indicated with following formula:
Wherein, D indicates the NARF key point range error, (xi,yi,zi) and (xi′,yi′,zi') it is respectively tentatively to match
The space coordinate of NARF key point, d are corresponded in two panels point cloud after standardiIndicate the distance between two corresponding NARF key points, n
Indicate the quantity of sampling NARF key point, summation symbol ∑ indicates the distance to sampling NARF key points pair all in two panels point cloud
Square summation.
The NARF key point range error is Euclidean distance between NARF key point is corresponded in first registration result equal
Root, rather than the Euclidean distance mean value between corresponding NARF key point；It is both needed to calculate the value due to being registrated iteration every time, and
Point cloud substantial amounts, such design can reduce the number of square root calculating to the full extent, while can ensure correspondence again
Euclidean distance mean value between NARF key point can be also well controlled.
On the basis of the above embodiments, the embodiment of the present invention is realized pair by introducing NARF key point range error
The effective evaluation of registration effect；NARF key point range error is smaller, and the matching precision between corresponding interior point is higher, puts matching for cloud
Quasi- effect is better.
Further, based on the above embodiment, the value of the just registration parameter includes: that the space support region is
0.05；The minimum interest value is 0.4, and the preset step-length that the minimum interest value is successively decreased is 0.025；The described search radius of neighbourhood
It is 0.07, the preset step-length of described search field increasing radii is 0.01, and described search field radius preset threshold is 0.15.
The initial value of the default just registration parameter is obtained according to a large amount of fruit tree three-dimensional point cloud with quasi-experiment.It is sent out through experiment
Existing, the point cloud under different shooting distances and angle, which needs to set different registration parameters, could obtain optimal registration effect.
In order to realize automation registration, have studied through a large number of experiments how initiation parameter and parameter variation range.
The size of the space support region of certain point determines NARF key point extracts whether have stability and difference in point cloud
Property, its value is positively correlated with calculating NARF key point the spent time when space support region is in 0.05~0.3 section, experiment hair
Present space support region is to spend the time minimum at 0.04 and 0.05, but since the value of support region is smaller, NARF key point is more inclined
To inflection point, therefore in order to reduce the influence of noise, while the stability for extracting NARF key point is improved, space support region is preset as
0.05。
Minimum interest value is that the decisive indicator of NARF key point is chosen from interior points close to the boundary, and interior points close to the boundary interest value is greater than
Minimum interest value is chosen as NARF key point.When one timing of support region, minimum interest value is bigger, and NARF key points are fewer.
Experiment discovery when minimum interest value is 0.4, can stablize the NARF key points for obtaining the larger value and the NARF key point concentrates packet
Containing most stability and distinctive NARF key point, meanwhile, minimum interest value often increaseds or decreases within this range
0.025, NARF key points will accordingly be reduced or increased about 1/3, therefore default minimum interest value decrement step size is 0.025；
When NARF key points are not up to expectation index, minimum interest value can be successively decreased by 0.025 until meet demand.
Whether Size of Neighborhood described in field search radius, that is, FPFH characteristic value, it is reasonable to choose, and determines that FPFH is retouched
State son whether can the geometrical characteristic accurately to local space where NARF key point be described and distinguish.Shoot different views
It is not prescribed by the distance between Kinect sensor and tree crown when the top fruit sprayer of angle, leads to the top fruit sprayer got point Yun Tiji
There are certain to zoom in or out, since the induction range of Kinect sensor is between 0.4~3 meter, so theoretically feature
The best search radius of value can also float in certain section.Experiment discovery, when in space, support region is 0.05, best field is searched
Rope radius is within the scope of 0.07~0.15, when according to 0.01 incremental search of step-length, can comparatively fast find ideal NARF key point
Several optimal values, therefore the initial value of field search radius is preset as 0.07, threshold preset 0.15, incremental steps are preset as
0.01。
On the basis of the above embodiments, the embodiment of the present invention passes through between the apparent different parameter presets of many experiments
Interaction rule has simultaneously determined best just registration parameter preset value, solves and manually adjusts time-consuming and laborious, mesh existing for parameter
The problems such as effect application condition is big is surveyed, the realization for automating registration Algorithm for cloud is laid a good foundation.Suitable space support region
The quantity for extracting NARF key point can be ensured with minimum interest value, and is ensured in the NARF set of keypoints extracted comprising point
Most stability and distinctive point set in cloud data reach relatively good registration effect, improve the efficiency of just registration process；
Neighborhood search radius preset value give space geometry local feature search initial value and search range, for be quickly obtained compared with
Support is provided for ideal NARF set of keypoints.
Further, based on the above embodiment, described to be iterated according to preset rules, optimize the registration parameter, has
Body includes: construction registration effect evaluation index Array for structural body, and presetting the NARF key point Duplication is 150, according to default
Step iteration reduces the minimum interest value, until meeting default NARF key point Duplication；It is incrementally increased according to preset step-length
During which described search field radius constantly obtains to preset threshold and purifies matching relationship using the purification match parameter, obtain
Multiple space conversion models and its NARF key point dispersion ratio and NARF key point range error are obtained, compares screening and obtains optimal institute
Just registration model is stated, and obtains the registration parameter optimized under the just registration model.
Registration effect evaluation index structural body is constructed, is specifically included: space support region, minimum interest value, field search half
Diameter, the range deviation threshold value of NARF key point feature corresponding relationship, NARF key point dispersion, NARF key point range error,
Space conversion matrix, most imperial palace points.
It is found through experiments that, when the minimum NARF key point in two panels point cloud reaches 150 or more, source point cloud carries out space
It is higher with the overlapping degree of target point cloud same section after transformation, relatively good first registration effect can be reached.
It is corresponding by parameter preset initialization space support region, minimum interest value, field search radius, NARF key point feature
The range deviation threshold value of relationship, search NARF key point simultaneously record keypoint quantity, if NARF keypoint quantity be not up to 150 with
On then minimum interest value successively decrease by 0.025 until two panels point cloud is all satisfied keypoint quantity demand.It is incremented by field by step-length 0.01
Search radius calculates two panels point cloud NARF key point characteristic value under each field search radius, and using most up to reaching 0.15
Nearest neighbor algorithm matches the NARF key point of source point cloud and target point cloud, is closed using the purification NARF key point matching of RANSAC algorithm
System removes repeated matching, obtains NARF key point space conversion models, counts and meet the NARF key point feature under the model
The quantity of the interior point of the range deviation threshold condition of corresponding relationship, is obtained under each neighborhood search radius most by iteration for several times
Imperial palace points and the space conversion models corresponding to it, and calculate its interior Y-coordinate very poor and Z coordinate it is very poor it
With, calculate in put range error, create the registration effect evaluation index structural body, and by most imperial palace points and its corresponding
Related data is saved in the registration effect evaluation index Array for structural body.
Registration effect evaluation index Array for structural body is traversed, optimal just registration effect is selected.NARF is chosen in array to close
The maximum subset of key point dispersion, and the smallest array element of NARF key point range error, the element are chosen in the subset
Corresponding space conversion matrix is optimal just registration model.
On the basis of the above embodiments, the embodiment of the present invention is by constructing the registration effect evaluation index structural body structure
The appraisement system of point cloud registering effect has been built, point cloud registering effect has scientifically and rationally been quantified, has realized and chooses optimal match automatically
The point cloud matching of quasi- effect.Parameter preset optimization, the NARF are realized by presetting the NARF key point Duplication index
Key point Duplication again simultaneously ensured enough interior numbers, number is more, represent source point cloud carry out spatial alternation after with mesh
The overlapping degree of punctuate cloud same section is bigger；The key point dispersion ratio be NARF key point matching relationship purification after interior point it
Between dispersion, dispersion is bigger, can more show a structure for cloud entirety, and point cloud registering effect is better；The key point away from
It is the Euclidean distance corresponded in first registration result between key point from error, range error is smaller, matches essence between corresponding interior point
Degree is higher, and registration effect is better.By be compared according to scientific and reasonable pre-defined rule and it is fast automatic screening obtain most
The first registration model of the point cloud of excellent registration effect realizes two panels point cloud automation registration.
Further, based on the above embodiment, the purification matching relationship is realized using RANSAC method；The purification
It is 0.02 with the range deviation threshold value described in parameter for purifying and matching NARF key point feature corresponding relationship.
RANSAC algorithm is found in a large amount of samples comprising noise data comprising at most interior point most by iterative manner
Excellent parameter model.Three corresponding relationships for randomly selecting NARF key point centering, calculate spatial mapping matrix and to source NARF
Key point point cloud carries out space reflection and obtains it in the coordinate in target point cloud space, compares the source NARF key point after mapping and its
The range deviation of corresponding target NARF key point in corresponding relationship, statistics meet the distance of NARF key point feature corresponding relationship
The point of deviation threshold condition is to logarithm, and as interior points, continuous iteration is to theoretical the number of iterations or maximum number of iterations is met, i.e.,
The most imperial palace points amount of can get and its corresponding space reflection model.Experiment shows in NARF key point feature corresponding relationship
When range deviation threshold value is set as 0.02, the ideal first registration model of interior points, and greatest iteration at this time can be quickly obtained
Number is set as 1000, and theoretical the number of iterations indicates are as follows:
Wherein, k is theoretical the number of iterations, which ensure that can obtain in the case where confidence level p (being set as 0.99)
Correct model parameter, t are interior number, and M is point cloud sample data, and m is the minimum samples randomly selected,It indicates
Random sampling is all probability at interior.
On the basis of the above embodiments, the embodiment of the present invention can have by using RANSAC purification matching relationship algorithm
The erroneous matching for inhibiting to effect the NARF key point feature as caused by the reasons such as noise, rejects erroneous matching, and purification matching is closed
System improves the spatial alternation correctness solved between two panels point cloud, obtains ideal first registration result.NARF key point
The reasonable setting of the range deviation threshold value of feature corresponding relationship had both ensured that the matching of NARF key point had degree of precision, also simultaneously
It ensure that algorithm has preferable efficiency.
Further, based on the above embodiment, described that the source point cloud obtained after first registration is iterated with target point cloud
Essence registration, specifically includes:
The first registration result after a cloud is just registrated is carried out according to using the just registration parameter of optimization, then is based on the essence
Registration parameter is iterated accuracy registration using ICP algorithm；The default preferentially parameter refers to preset for smart registration point cloud
Maximum corresponding points distance and range error change threshold；The range error variation of the essence registration point cloud refers to adjacent twice change
The difference of the range error of smart registration point cloud after generation registration；
The maximum corresponding points distance of the essence registration point cloud is preset as 0.07；The range error variation of the essence registration point cloud
Threshold preset is 10-8；
The corresponding points distance of the essence registration point cloud is indicated using following formula:
Wherein, (x, y, z) and (x ', y ', z ') is respectively the space coordinate of corresponding points in the two panels point cloud after essence is registrated, d
Indicate the distance between two corresponding points；
The calculation method of the range error variation of the essence registration point cloud is as follows:
V=| Lm-Lm-1|
Wherein, L indicates the range error of two panels essence registration point cloud, (xl,yl,zl) and (xl′,yl′,zl') it is respectively accurate
The space coordinate of corresponding points, d in two panels point cloud after registrationlIndicate the distance between two corresponding points, N indicates point cloud midpoint pair
Quantity, summation symbol ∑ indicate in two panels point cloud all corresponding points pair square distance sum；M indicates current iteration time
Number, v indicate the range error variation of two panels essence registration point cloud, by the m time iteration and the smart registration point cloud of the m-1 times iteration away from
Difference from error determines.
Preferable spatial relationship can be obtained by just putting after registration between cloud, and point tolerance can in the model that RANSAC is obtained at this time
To be maintained at 0.02 or less.The setting of maximum corresponding points distance is the corresponding points in order to be quickly found out in two panels point cloud, through testing
It was found that by the present embodiment just with will definitely for essence be registrated the two panels initial point cloud of ideal position be provided, when maximum corresponding points away from
From being preset as 0.07, the range error change threshold of smart registration point cloud is preset as 10-8When, it can be obtained by 2 ICP algorithm iteration
Obtain the point cloud registering result of degree of precision.
ICP algorithm is based on the source point cloud and target point cloud obtained after first registration, for each point in source point cloud, in target
The corresponding points composition point Cloud transform point that nearest and distance is less than the maximum respective distances of smart registration point cloud is found in point cloud
Collection；Rotational transformation matrix and translation transformation square of the point Cloud transform point concentration between source to target are calculated using singular value decomposition method
Battle array, and carry out from source point cloud to the transformation of target point cloud；Carry out calculating after iteration twice according to the above method smart registration point cloud away from
Difference from error, the range error change threshold if more than smart registration point cloud then continue iteration, until meeting smart registration point cloud
Range error change threshold condition until.
On the basis of the above embodiments, the reasonable setting that the embodiment of the present invention passes through maximum corresponding points distance effectively reduces
Wrong corresponding points in smart registration process improve registration accuracy and with Quasi velosity.The range error of smart registration point cloud changes
Threshold value is smaller, indicates that the registration accuracy for the smart registration model for meeting the requirement is higher.Meeting the default preferentially parameter of two above
When, ICP algorithm stops iteration, obtains two panels point cloud essence registration result, realizes the two panels point cloud automation registration of degree of precision.
Fig. 2 is top fruit sprayer three-dimensional point cloud Automated Refactoring flow chart provided in an embodiment of the present invention.Such as Fig. 2 institute
Show, which comprises
Step 201 automates method for registering to adjacent view top fruit sprayer three-dimensional point cloud using top fruit sprayer three-dimensional point cloud
Essence registration two-by-two saves intermediate conversion matrix, is in certain sequence transformed into all the points cloud most by the intermediate conversion matrix
Space coordinates where the point cloud of the latter visual angle, obtain top fruit sprayer three-dimensional point cloud；The adjacent view point cloud, refers to use
Kinect acquires the consecutive points cloud that the two panels point cloud Duplication obtained during fruit tree three dimensional point cloud is not less than 30%；It is described
Intermediate conversion matrix refers to the registration transition matrix from source point cloud spatial alternation to target point cloud space.
Wherein, the top fruit sprayer three-dimensional point cloud automation method for registering can be what any embodiment of the present invention was proposed
Top fruit sprayer three-dimensional point cloud automates method for registering.
Experiment discovery can ensure adjacent two panels point when the adjacent two panels point cloud Duplication of top fruit sprayer is not less than 30%
In cloud any point cloud have enough NARF key point Duplication, be two panels point cloud accuracy registration and entire tree crown it is accurate
Registration provides data basis.
When top fruit sprayer accuracy registration, by point cloud serial number, if the 1st point cloud is the last one visual angle point cloud.By the 2nd
Piece point cloud is retained from the 2nd point cloud to the 1st point cloud by being just registrated and smart registration process is matched to the 1st point cloud space
Transformation matrix；3rd point cloud is passed through into just registration and smart registration transformation to the 2nd point cloud space first, is retained from the 3rd point cloud
The 3rd point cloud is passed through from the 2nd point cloud to the 1st point to the transformation matrix of the 2nd point cloud, then from the 2nd point cloud space
The transformation matrix of cloud transforms to the 1st point cloud space.And so on, until first visual angle point Cloud transform is empty to the 1st point cloud
Between until.
On the basis of the above embodiments, the precondition that the embodiment of the present invention passes through proposition adjacent view point cloud Duplication
The good accuracy registration of two panels point cloud and entire tree crown is ensured.By adjacent view point cloud after accuracy registration two-by-two, protect
Intermediate conversion matrix is deposited, all the points cloud is finally transformed into the last one view by multiple intermediate conversion matrixes in sequence
Space coordinates where angle point cloud have ensured that matching error is evenly dispersed in the point cloud at each visual angle, reduce repeatedly
Error accumulation caused by iteration realizes accurate top fruit sprayer point cloud registering.
Further, based on the above embodiment, it is described by the intermediate conversion matrix in certain sequence by certain piece point cloud
Be transformed into point cloud accuracy registration transformation matrix used in the space coordinates where the last one visual angle point cloud is indicated with following formula:
Wherein, F indicates described cloud accuracy registration transformation matrix, Mi,i-1It indicates from i-th cloud space to (i-1)-th cloud sky
Between intermediate conversion matrix, n indicates to select cloud from certain visual angle and orders cloud sheet number to what cloud was selected at a last visual angle, and quadrature symbol ∏ is indicated
To intermediate conversion Matrix Calculating products all in registration process.
When being transformed into (i-1)-th cloud space from i-th cloud space, the i-th spatial point cloud is transformed into (i-1)-th cloud first
Space, i.e., by being just registrated and the process of essence registration is by i-th point Cloud transform to (i-1)-th point cloud space, M at this timei,i-1As
Its transformation matrix.Similarly, (i-1)-th point cloud needs to be converted into the i-th -2 point cloud spaces by same step, it is converted at this time
Matrix is Mi-1,i-2.Then i-th point cloud can be by transformation matrix Mi-1,i-2It is converted into the i-th -2 point cloud spaces, Mi,i-1×
Mi-1,i-2For its transformation matrix.And so on, until by i-th point Cloud transform to the 1st point cloud space, transformation matrix
As F.
On the basis of the above embodiments, the embodiment of the present invention realizes any spatial point by designing F iteration expression formula
Cloud to final goal space matched transform automation algorithm, for top fruit sprayer point cloud matching go on smoothly provide it is efficient
Algorithm guarantee.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer readable storage medium, the program
When being executed, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes: ROM, RAM, magnetic disk or light
The various media that can store program code such as disk.
The embodiments such as computer equipment described above are only schematical, wherein described be used as separate part description
Unit may or may not be physically separated, component shown as a unit may or may not be
Physical unit, it can it is in one place, or may be distributed over multiple network units.It can be according to the actual needs
Some or all of the modules therein is selected to achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying
In the case where creative labor, it can understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should
Computer software product may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features；
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (9)
1. a kind of top fruit sprayer three-dimensional point cloud automates method for registering characterized by comprising
Default registration parameter, is iterated according to preset rules, optimizes the registration parameter, utilize the registration parameter of optimization
Carry out point cloud registering；The registration parameter includes first registration parameter, purification match parameter and smart registration parameter；Wherein, described first
Registration parameter is for determining the minimum interest value of NARF key point, space support region and for calculating NARF key point
The search field radius of FPFH characteristic value, the purification match parameter are for purifying and matching the corresponding pass of NARF key point feature
The range deviation threshold value of system, the essence registration parameter are maximum corresponding points distance and the range error variation for smart registration point cloud
Threshold value；
The preset rules are registration effect evaluation index, and according to priority orders, the registration effect evaluation index includes
NARF key point Duplication, NARF key point dispersion ratio and NARF key point range error；The NARF key point Duplication is
Ensure the NARF key point number after matching relationship purifies with point quantity in enough；The NARF key point dispersion ratio is NARF
Dispersion after the purification of key point matching relationship between interior point；The NARF key point range error is right in preliminary registration result
Answer the root mean square of the Euclidean distance between NARF key point；Wherein, the interior point refers in just registration process according to just with quasi-mode
Calculated range error meets the NARF key point of preset condition after type matching；
To the source point cloud obtained after first registration and target point cloud be iterated it is smart be registrated, when meeting default preferentially parameter, stopping
Iteration obtains two panels point cloud essence registration result.
2. the method according to claim 1, wherein the NARF key point dispersion ratio chooses NARF key point y
The the sum of very poor of the very poor and z coordinate of coordinate indicates:
S=| ymax-ymin|+|zmax-zmin|
Wherein, s indicates the NARF key point dispersion ratio, ymaxAnd yminThe respectively maximum value and minimum value of y-coordinate axis direction,
zmaxAnd zminThe respectively maximum value and minimum value of z coordinate axis direction.
3. the method according to claim 1, wherein the NARF key point range error is indicated with following formula:
Wherein, D indicates the NARF key point range error, (xi,yi,zi) and (xi′,yi′,zi') be respectively preliminary registration after
Two panels point cloud in correspond to the space coordinate of NARF key point, diIndicate the distance between two corresponding NARF key points, n is indicated
The quantity of NARF key point is sampled, summation symbol ∑ indicates the square distance to sampling NARF key points pair all in two panels point cloud
Summation.
4. the method according to claim 1, wherein the value of the just registration parameter includes: the space branch
Holding domain is 0.05；The minimum interest value is 0.4, and the preset step-length that the minimum interest value is successively decreased is 0.025；Described search is adjacent
Domain radius is 0.07, and the preset step-length of described search field increasing radii is 0.01, and described search field radius preset threshold is
0.15。
5. matching described in optimization the method according to claim 1, wherein described be iterated according to preset rules
Quasi- parameter, specifically includes: construction registration effect evaluation index Array for structural body, presetting the NARF key point Duplication is 150,
Reduce the minimum interest value according to preset step-length iteration, until meeting default NARF key point Duplication；According to preset step-length
Described search field radius is incrementally increased to preset threshold, during which constantly obtains and utilizes purification match parameter purification matching
Relationship obtains multiple space conversion models and its NARF key point dispersion ratio and NARF key point range error, compares and screen
To the optimal just registration model, and obtain the registration parameter optimized under the just registration model.
6. the method according to claim 1, wherein the purification matching relationship is realized using RANSAC method；
Range deviation threshold value described in the purification match parameter for purifying and matching NARF key point feature corresponding relationship is
0.02。
7. the method according to claim 1, wherein described to the source point cloud and target point cloud that are obtained after first registration
It is iterated smart registration, is specifically included:
The first registration result after a cloud is just registrated is carried out according to the registration parameter using optimization, then is preset preferentially based on described
Parameter is iterated accuracy registration using ICP algorithm；The default preferentially parameter refer to it is preset for smart registration point cloud most
Big corresponding points distance and range error change threshold；The range error variation of the essence registration point cloud refers to that adjacent iteration is matched twice
The difference of the range error of smart registration point cloud after quasi-；
The maximum corresponding points distance of the essence registration point cloud is preset as 0.07；The range error change threshold of the essence registration point cloud
It is preset as 10-8；
The corresponding points distance of the essence registration point cloud is indicated using following formula:
Wherein, (x, y, z) and (x ', y ', z ') is respectively the space coordinate of corresponding points in the two panels point cloud after essence is registrated, and d is indicated
The distance between two corresponding points；
The calculation method of the range error variation of the essence registration point cloud is as follows:
V=| Lm-Lm-1|
Wherein, L indicates the range error of two panels essence registration point cloud, (xl,yl,zl) and (xl′,yl′,zl') it is respectively accuracy registration
The space coordinate of corresponding points, d in two panels point cloud afterwardslIndicate the distance between two corresponding points, N indicates the number at point cloud midpoint pair
Amount, summation symbol ∑ indicate to sum to the square distance of all corresponding points pair in two panels point cloud；M indicates current iteration number, v table
The range error variation for showing two panels essence registration point cloud, by the range error of the smart registration point cloud of the m times iteration and the m-1 times iteration
Difference determine.
8. a kind of top fruit sprayer three-dimensional point cloud Automated Refactoring based on any the method for claim 1 to 7, feature
It is, comprising:
Using any top fruit sprayer three-dimensional point cloud automation method for registering of claim 1~7 to adjacent view fruit tree tree
It is preced with three-dimensional point cloud essence registration two-by-two, intermediate conversion matrix is saved, by the intermediate conversion matrix in certain sequence by all the points
Cloud is transformed into the space coordinates where the last one visual angle point cloud, obtains top fruit sprayer three-dimensional point cloud；The adjacent view point
Cloud refers to using phase of the two panels point cloud Duplication not less than 30% obtained during Kinect acquisition fruit tree three dimensional point cloud
Adjoint point cloud；The intermediate conversion matrix refers to the registration transition matrix from source point cloud spatial alternation to target point cloud space.
9. according to the method described in claim 8, it is characterized in that, described in certain sequence will by the intermediate conversion matrix
Certain piece point cloud is transformed under point cloud accuracy registration transformation matrix used in the space coordinates where the last one visual angle point cloud uses
Formula indicates:
Wherein, F indicates i-th point cloud accuracy registration transformation matrix, Mi,i-1It indicates from i-th cloud space to (i-1)-th cloud
The intermediate conversion matrix in space, n indicate to select cloud from certain visual angle and order cloud sheet number to what cloud was selected at a last visual angle, quadrature symbol ∏ table
Show to intermediate conversion Matrix Calculating products all in registration process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811558734.6A CN109801316B (en) | 2018-12-19 | 2018-12-19 | Automatic registration method and reconstruction method for three-dimensional point cloud of crown of fruit tree |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811558734.6A CN109801316B (en) | 2018-12-19 | 2018-12-19 | Automatic registration method and reconstruction method for three-dimensional point cloud of crown of fruit tree |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109801316A true CN109801316A (en) | 2019-05-24 |
| CN109801316B CN109801316B (en) | 2021-02-23 |
Family
ID=66557319
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811558734.6A Active CN109801316B (en) | 2018-12-19 | 2018-12-19 | Automatic registration method and reconstruction method for three-dimensional point cloud of crown of fruit tree |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109801316B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110766733A (en) * | 2019-10-28 | 2020-02-07 | 广东三维家信息科技有限公司 | Single-space point cloud registration method and device |
| CN111009029A (en) * | 2019-11-25 | 2020-04-14 | 视辰信息科技(上海)有限公司 | Data processing method and device for three-dimensional reconstruction, electronic equipment and storage medium |
| CN111369603A (en) * | 2020-02-25 | 2020-07-03 | 北京百度网讯科技有限公司 | Point cloud registration technology evaluation method and device, electronic equipment and readable storage medium |
| CN111612909A (en) * | 2020-05-14 | 2020-09-01 | 北京林业大学 | Three-dimensional branch breaking effect simulation algorithm based on point cloud data |
| CN111738293A (en) * | 2020-05-18 | 2020-10-02 | 北京百度网讯科技有限公司 | Point cloud data processing method and device, electronic equipment and readable storage medium |
| CN111815706A (en) * | 2020-06-23 | 2020-10-23 | 熵智科技(深圳)有限公司 | Visual identification method, device, equipment and medium for single-article unstacking |
| CN112802071A (en) * | 2021-01-22 | 2021-05-14 | 北京农业智能装备技术研究中心 | Three-dimensional reconstruction effect evaluation method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104143210A (en) * | 2014-07-31 | 2014-11-12 | 哈尔滨工程大学 | Multi-scale normal feature point cloud registering method |
| CN108537732A (en) * | 2018-04-10 | 2018-09-14 | 福州大学 | Fast image splicing method based on PCA-SIFT |
| CN108564605A (en) * | 2018-04-09 | 2018-09-21 | 大连理工大学 | A kind of three-dimensional measurement spots cloud optimization method for registering |
-
2018
- 2018-12-19 CN CN201811558734.6A patent/CN109801316B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104143210A (en) * | 2014-07-31 | 2014-11-12 | 哈尔滨工程大学 | Multi-scale normal feature point cloud registering method |
| CN108564605A (en) * | 2018-04-09 | 2018-09-21 | 大连理工大学 | A kind of three-dimensional measurement spots cloud optimization method for registering |
| CN108537732A (en) * | 2018-04-10 | 2018-09-14 | 福州大学 | Fast image splicing method based on PCA-SIFT |
Non-Patent Citations (3)
| Title |
|---|
| MAI CHUNYAN: "Rapid 3D reconstruction of fruit tree based on point", 《农业工程学报》 * |
| 贾勇杰: "基于SHOT的多尺度关键点检测技术研究", 《激光与光电子学进展》 * |
| 郑立华: "基于Kinect相机的苹果树三维点云配准", 《农业机械学报》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110766733A (en) * | 2019-10-28 | 2020-02-07 | 广东三维家信息科技有限公司 | Single-space point cloud registration method and device |
| CN110766733B (en) * | 2019-10-28 | 2022-08-12 | 广东三维家信息科技有限公司 | Single-space point cloud registration method and device |
| CN111009029A (en) * | 2019-11-25 | 2020-04-14 | 视辰信息科技(上海)有限公司 | Data processing method and device for three-dimensional reconstruction, electronic equipment and storage medium |
| CN111009029B (en) * | 2019-11-25 | 2021-05-11 | 视辰信息科技(上海)有限公司 | Data processing method and device for three-dimensional reconstruction, electronic equipment and storage medium |
| CN111369603A (en) * | 2020-02-25 | 2020-07-03 | 北京百度网讯科技有限公司 | Point cloud registration technology evaluation method and device, electronic equipment and readable storage medium |
| CN111369603B (en) * | 2020-02-25 | 2023-09-29 | 阿波罗智能技术(北京)有限公司 | Evaluation method and device of point cloud registration technology, electronic equipment and readable storage medium |
| CN111612909A (en) * | 2020-05-14 | 2020-09-01 | 北京林业大学 | Three-dimensional branch breaking effect simulation algorithm based on point cloud data |
| CN111738293A (en) * | 2020-05-18 | 2020-10-02 | 北京百度网讯科技有限公司 | Point cloud data processing method and device, electronic equipment and readable storage medium |
| CN111815706A (en) * | 2020-06-23 | 2020-10-23 | 熵智科技(深圳)有限公司 | Visual identification method, device, equipment and medium for single-article unstacking |
| CN111815706B (en) * | 2020-06-23 | 2023-10-27 | 熵智科技(深圳)有限公司 | Visual identification method, device, equipment and medium for single-item unstacking |
| CN112802071A (en) * | 2021-01-22 | 2021-05-14 | 北京农业智能装备技术研究中心 | Three-dimensional reconstruction effect evaluation method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109801316B (en) | 2021-02-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109801316A (en) | A kind of top fruit sprayer three-dimensional point cloud automation method for registering and reconstructing method | |
| Huang et al. | Shape synthesis from sketches via procedural models and convolutional networks | |
| Pound et al. | Automated recovery of three-dimensional models of plant shoots from multiple color images | |
| CN110340891B (en) | Mechanical arm positioning and grabbing system and method based on point cloud template matching technology | |
| CN107038717B (en) | A method of 3D point cloud registration error is automatically analyzed based on three-dimensional grid | |
| Gibbs et al. | Approaches to three-dimensional reconstruction of plant shoot topology and geometry | |
| CN106529469B (en) | Unmanned aerial vehicle-mounted LiDAR point cloud filtering method based on self-adaptive gradient | |
| JP4924165B2 (en) | Shape comparison device and shape comparison method | |
| Gibbs et al. | Plant phenotyping: an active vision cell for three-dimensional plant shoot reconstruction | |
| CN108171780A (en) | A kind of method that indoor true three-dimension map is built based on laser radar | |
| WO2015188445A1 (en) | Point cloud three-dimensional model reconstruction method and system | |
| CN111862289B (en) | Point cloud up-sampling method based on GAN network | |
| CN110942515A (en) | Point cloud-based target object three-dimensional computer modeling method and target identification method | |
| CN110490917A (en) | Three-dimensional rebuilding method and device | |
| CN110992473A (en) | Tree branch modeling method and system based on vehicle-mounted laser scanning point cloud | |
| Li et al. | Dictionary optimization and constraint neighbor embedding-based dictionary mapping for superdimension reconstruction of porous media | |
| CN114510772A (en) | Rapid generation method of building surface contour based on oblique scanning data | |
| CN113450269A (en) | Point cloud key point extraction method based on 3D vision | |
| CN114663373A (en) | Point cloud registration method and device for detecting surface quality of part | |
| CN106611030A (en) | Object similarity comparison method and object search method based on video, and object similarity comparison system and object search system based on video | |
| Pan et al. | A global generalized maximum coverage-based solution to the non-model-based view planning problem for object reconstruction | |
| Golla et al. | Temporal upsampling of point cloud sequences by optimal transport for plant growth visualization | |
| CN116704137B (en) | Reverse modeling method for point cloud deep learning of offshore oil drilling platform | |
| FI121981B (en) | Following an object in machine vision | |
| CN117011465A (en) | Tree three-dimensional reconstruction method and device, electronic equipment and storage medium |
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 |