CN106250881A - A kind of target identification method based on three dimensional point cloud and system - Google Patents

A kind of target identification method based on three dimensional point cloud and system Download PDF

Info

Publication number
CN106250881A
CN106250881A CN201610727765.4A CN201610727765A CN106250881A CN 106250881 A CN106250881 A CN 106250881A CN 201610727765 A CN201610727765 A CN 201610727765A CN 106250881 A CN106250881 A CN 106250881A
Authority
CN
China
Prior art keywords
cloud
point cloud
dimensional point
identified
target
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.)
Pending
Application number
CN201610727765.4A
Other languages
Chinese (zh)
Inventor
田劲东
单波
田勇
李东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen University
Original Assignee
Shenzhen University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen University filed Critical Shenzhen University
Priority to CN201610727765.4A priority Critical patent/CN106250881A/en
Publication of CN106250881A publication Critical patent/CN106250881A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/60Type of objects
    • G06V20/64Three-dimensional objects
    • G06V20/653Three-dimensional objects by matching three-dimensional models, e.g. conformal mapping of Riemann surfaces
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/51Indexing; Data structures therefor; Storage structures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/22Matching criteria, e.g. proximity measures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/40Extraction of image or video features
    • G06V10/44Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Library & Information Science (AREA)
  • Processing Or Creating Images (AREA)

Abstract

The invention discloses a kind of target identification method based on three dimensional point cloud, it comprises the following steps: three dimensional point cloud other to marking class carries out circumscribed rectangular body to be asked for;Extracting rule solid from described marking class other three dimensional point cloud;Build model database;Build model data table;Input the cloud data of target to be identified and carry out ONLINE RECOGNITION.A kind of target identification system based on three dimensional point cloud, including: the first module, carry out circumscribed rectangular body for execution three dimensional point cloud other to marking class and ask for;Second module, for performing extracting rule solid from described marking class other three dimensional point cloud;Three module, is used for performing to build model database;4th module, is used for performing to build model data table;5th module, for performing input the cloud data of target to be identified and carry out ONLINE RECOGNITION.The method and system carry out during target recognition convenient and swift, and accuracy is higher.It is widely used in mode identification technology.

Description

A kind of target identification method based on three dimensional point cloud and system
Technical field
The present invention relates to mode identification technology, be specifically related to the three dimensional point cloud of a kind of rule-based solid Target identification method and system.
Background technology
Along with developing rapidly of 3-D scanning technology, target recognition based on three dimensional point cloud just wide by people General concern.Target recognition in three dimensional point cloud generally comprises two stages: feature extraction and characteristic matching.Object features bag Including global characteristics and local feature, what global characteristics described is the Global shape feature of object;And local feature is more focused on carefully Joint, description is the feature more become more meticulous in the little scope of object.Objective extraction based on global characteristics is to target deformation, mesh Mark because blocking the factor such as incomplete the most sensitive;And some local features have the advantages such as yardstick and rotational invariance, energy Enough solve the problems referred to above to a certain extent.Therefore, carrying out target recognition based on local feature is that current application compares widely Method.The method of characteristic matching can also be divided into two kinds: directly characteristic point matching method and indirect characteristic point matching method.Directly Characteristic point matching method directly calculates the matching degree of extracted feature on model with object to be identified, by statistical match success Characteristic point percentage ratio be identified;Characteristic point matching method is by the feature in model is carried out further group indirectly Close, it is achieved by the local feature description to global characteristics, between model and target to be matched, then carry out these feature groups The coupling closed carries out target recognition.Stable local feature combination is suitable for direct Feature Points Matching, energy in some scenes Enough obtain good effect, but utilize limited local feature to realize the description of the global characteristics to object and remain in complex scene The difficult point of target recognition.
Summary of the invention
In order to solve above-mentioned technical problem, it is an object of the invention to provide a kind of target recognition based on three dimensional point cloud Method and system.
The technical solution adopted in the present invention is:
The present invention provides a kind of target identification method based on three dimensional point cloud, and it comprises the following steps:
Three dimensional point cloud other to marking class carries out circumscribed rectangular body to be asked for;
Extracting rule solid from described marking class other three dimensional point cloud;
Build model database;
Build model data table;
Input the cloud data of target to be identified and carry out ONLINE RECOGNITION.
As the improvement of this technical scheme, described step three dimensional point cloud other to marking class carries out circumscribed rectangular body and asks Taking, it also includes:
Ask for described some cloud circumscribed rectangular body each crest line direction;
Ask for described some cloud circumscribed rectangular body.
As the improvement of this technical scheme, described regular geometric body includes cuboid and spheroid.
Further, described step is extracting rule solid from described marking class other three dimensional point cloud, also includes:
Use uniform grid method that other some cloud mass of described each marking class is carried out point cloud compressing;
Utilization solves the eigenvalue of sampled point nearest-neighbor point covariance matrix and the method for the normal vector of characteristic vector, right An obtained cloud of simplifying carries out normal vector estimation;
Use curvature method for solving based on local surface, an obtained cloud of simplifying is carried out Curvature Estimate;
Gained is simplified a cloud, according to the some cloud normal vector asked for and curvature, extracts the boundary characteristic of three-dimensional point cloud;
The cloud data with boundary characteristic extracted is carried out edge fitting, and judges that the solid simulated is No for regular geometric body;The most then estimate the canonical parameter of regular geometric body;Not, then this step is terminated.
Further, described step builds model database, comprising:
Gather the cloud data of each object to be identified, obtain the other multiple three dimensional point cloud of marking class;
Respectively the cloud data gathered is simplified;
Ask for the circumscribed rectangular body of each point cloud model and put the regular geometric body in cloud, building model database.
Further, described step inputs the cloud data of target to be identified and carries out ONLINE RECOGNITION, comprising:
The cloud data inputted is carried out cluster segmentation, obtains the some cloud mass to be identified of non-band classification information;
Ask for the circumscribed rectangular body of the cloud data of described to be identified some cloud mass, obtain the spy to be identified of non-band classification information Levy description;
Extract the regular geometric body in described cloud data, obtain the Feature Descriptor to be identified of non-band classification information;
According to constructed model data table and the Feature Descriptor of calculating, mate, it is achieved the knowledge of cloud data Not.
On the other hand, the present invention also provides for a kind of target identification system based on three dimensional point cloud, including:
First module, carries out circumscribed rectangular body for execution three dimensional point cloud other to marking class and asks for;
Second module, for performing extracting rule solid from described marking class other three dimensional point cloud;
Three module, is used for performing to build model database;
4th module, is used for performing to build model data table;
5th module, for performing input the cloud data of target to be identified and carry out ONLINE RECOGNITION.
The invention has the beneficial effects as follows: the present invention provide a kind of based on three dimensional point cloud target identification method and be System, by a cloud sample carries out pretreatment, then carries out feature extraction, estimation point cloud Feature Descriptor, structure to a cloud sample Established model data base, according to dependency, by the some cloud Feature Descriptor of target to be identified with in the model database built Point cloud Feature Descriptor mates, thus realizes the target recognition of three dimensional point cloud.The method and system carry out target knowledge Time other convenient and swift, and accuracy is higher.
Accompanying drawing explanation
Below in conjunction with the accompanying drawings the detailed description of the invention of the present invention is described further:
Fig. 1 is the schematic diagram of common regular geometric body cuboid in cloud data;
Fig. 2 is the schematic diagram of common regular geometric body spheroid in cloud data;
Fig. 3 is object circumscribed rectangular body schematic diagram;
Fig. 4 is the schematic flow sheet of target recognition one embodiment in three dimensional point cloud.
Detailed description of the invention
It should be noted that in the case of not conflicting, the embodiment in the application and the feature in embodiment can phases Combination mutually.
The present invention provides a kind of target identification method based on three dimensional point cloud, and it comprises the following steps:
Three dimensional point cloud other to marking class carries out circumscribed rectangular body to be asked for;
Extracting rule solid from described marking class other three dimensional point cloud;
Build model database;
Build model data table;
Input the cloud data of target to be identified and carry out ONLINE RECOGNITION.
As the improvement of this technical scheme, described step three dimensional point cloud other to marking class carries out circumscribed rectangular body and asks Taking, it also includes:
Ask for described some cloud circumscribed rectangular body each crest line direction;
Ask for described some cloud circumscribed rectangular body.
With reference to Fig. 1-2, as the improvement of this technical scheme, described regular geometric body includes cuboid and spheroid.
Further, described step is extracting rule solid from described marking class other three dimensional point cloud, also includes:
Use uniform grid method that other some cloud mass of described each marking class is carried out point cloud compressing;
Utilization solves the eigenvalue of sampled point nearest-neighbor point covariance matrix and the method for the normal vector of characteristic vector, right An obtained cloud of simplifying carries out normal vector estimation;
Use curvature method for solving based on local surface, an obtained cloud of simplifying is carried out Curvature Estimate;
Gained is simplified a cloud, according to the some cloud normal vector asked for and curvature, extracts the boundary characteristic of three-dimensional point cloud;
The cloud data with boundary characteristic extracted is carried out edge fitting, and judges that the solid simulated is No for regular geometric body;The most then estimate the canonical parameter of regular geometric body;Not, then this step is terminated.
Further, described step builds model database, comprising:
Gather the cloud data of each object to be identified, obtain the other multiple three dimensional point cloud of marking class;
Respectively the cloud data gathered is simplified;
Ask for the circumscribed rectangular body of each point cloud model, with reference to Fig. 3, be object circumscribed rectangular body schematic diagram, and in some cloud Regular geometric body, build model database.
Further, described step inputs the cloud data of target to be identified and carries out ONLINE RECOGNITION, comprising:
The cloud data inputted is carried out cluster segmentation, obtains the some cloud mass to be identified of non-band classification information;
Ask for the circumscribed rectangular body of the cloud data of described to be identified some cloud mass, obtain the spy to be identified of non-band classification information Levy description;
Extract the regular geometric body in described cloud data, obtain the Feature Descriptor to be identified of non-band classification information;
According to constructed model data table and the Feature Descriptor of calculating, mate, it is achieved the knowledge of cloud data Not.
On the other hand, the present invention also provides for a kind of target identification system based on three dimensional point cloud, including:
First module, carries out circumscribed rectangular body for execution three dimensional point cloud other to marking class and asks for;
Second module, for performing extracting rule solid from described marking class other three dimensional point cloud;
Three module, is used for performing to build model database;
4th module, is used for performing to build model data table;
5th module, for performing input the cloud data of target to be identified and carry out ONLINE RECOGNITION.
With reference to Fig. 4, it it is the schematic flow sheet of target recognition one embodiment in three dimensional point cloud.Comprising: to marking class Other three dimensional point cloud carries out circumscribed rectangular body to be asked for;
Extracting rule solid from described marking class other three dimensional point cloud;
Build model database;
Build model data table;
The cloud data inputting target to be identified is acquired, and analysis modeling, carry out ONLINE RECOGNITION.
Target identification method in the three dimensional point cloud that the present invention provides is divided into off-line training and two rank of ONLINE RECOGNITION Section.Off-line training process needs the some cloud sample of labelling in advance, first a cloud sample is carried out pretreatment, then enters a cloud sample Row feature extraction, estimation point cloud Feature Descriptor, build model database.Each sample can be obtained by three-dimensional scanning device, Target to be identified and sample use identical three-dimensional scanning device to obtain under similarity condition.The ONLINE RECOGNITION stage, according to phase Guan Xing, is carried out by some cloud Feature Descriptor and the some cloud Feature Descriptor in the model database built of target to be identified Join, thus realize the target recognition of three dimensional point cloud.This embodiment carries out discrete model instruction for different classes of target Practice.
Off-line training process, specifically includes following steps:
Step 1, three dimensional point cloud other to marking class carry out circumscribed rectangular body to be asked for, and wherein step specifically includes:
Described step 1 includes A11: ask for cloud circumscribed rectangular body each crest line direction.Specifically include following steps:
Q11, ask for two points that distance in cloud data is maximum, A and B.
Q12, point of contact A and B, obtain line segment AB, and solve the midpoint O of line segment AB.
Q13, excessively O point make the vertical line L1 of line segment AB.
Q14, straight line L1 excessively make plane M.
Q15, in plane M, distance value between points in estimation point cloud.Selected point line is through O point and distance value 2 maximum points, some C and some D, point of contact C and some D, obtain line segment CD.
Q16, excessively O point make the straight line L2 both perpendicular to straight line AB and straight line CD.
In described step 1, A12 asks for a cloud circumscribed rectangular body.Specifically include following steps:
Z11, it is cuboid crest line direction with straight line AB, straight line CD, straight line L2 respectively, long with different ribs, build cuboid.
Z12, when a cloud is contained by cuboid just, this cuboid is required cuboid.
Step 2, three dimensional point cloud other to marking class carry out feature extraction, specifically include:
A21 in described step 2: other some cloud mass of each marking class is carried out point cloud compressing, specifically includes following steps:
S11, in the plane vertical with scanning direction, set up uniform multiple plane grid.
S12, the data point in other for each marking class cloud mass is assigned in the grid that step S11 is set up.
These distance values are arranged by S13, the distance of calculating each point to grid plane according to order from big to small.
S14, for belonging to the point of same grid, selected distance value replaces whole grid equal to the data point of intermediate value Data point, thus realize point cloud compressing.
In described step 2, the A22 key point to obtaining in step A21 carries out normal vector and asks for, and uses directly from a cloud number According to middle estimation point cloud normal vector, eigenvalue and characteristic vector by sampled point nearest-neighbor point covariance matrix carry out normal vector Solve, specifically include following steps:
S21, the initial three-dimensional cloud data employing KD-Tree structure storage that will input, set radius of neighbourhood R=0.5m, To each point p in a cloudi(x, y z), calculate its radius neighborhood point set pk
S22, set space plane equation as ax+by+cz=d, wherein (a b c)TIt is the unit vector in plane normal direction, I.e. a2+b2+c2=1, d are the distance that initial point arrives this plane.To its radius neighborhood pkIn point set (xi, yi, zi), available matching Error:Wherein di=| axi+byi+czi-d|。
S23, seek error minima, obtain optimal planar, and set up Lagrange's equation:
S24, to above-mentioned formula is asked local derviation respectively, and to make local derviation numerical value be zero, i.e. Obtain equation A [a, b, c] '=λ [a, b, c].And minimal eigenvalue characteristic of correspondence vector is (a b c)T, thus estimate Normal vector.
In described step 2, the A23 key point to obtaining in step A21 carries out curvature and estimates, uses local surface Curvature method for solving, specifically includes following steps:
S31 is according to the implicit equation of dimensional conicoide
For three dimensional point cloud, method of least square is utilized to obtain through some P0And the optimum surface equation of neighborhood point, obtain The unit normal vector n of this point0(nx, ny, nz)。
S32 calculates the sagittal plane equation M through point1(x, y, z)=A1x+B1y+C1z+D1=0.
S33 Simultaneous EquationsTry to achieve quadratic surface and radial direction is bent The crossing space curve in face.
S34 is for space curve, and on curve, the curvature of a fixed point is
Curvature can be tried to achieve.
In described step 2, A24 is for point cloud boundary characteristics, utilizes the some cloud normal vector and song asked in step A22 and A23 Rate, it is achieved the Boundary characteristic extraction of three-dimensional point cloud.Specifically include following steps:
S41, utilize in step A22 the point cloud normal vector of estimation, calculate set point and the normal vector of each point in its k nearest neighbor Angle, and determine whether characteristic point candidate point by the size judging all of angle value and given threshold value.If each angle Angle value both greater than threshold value is then candidate feature point;Otherwise, it not candidate point and rejected.
S42, on the basis of step S41, set curvature threshold, by step A23 estimation principal curvatures and set threshold The size of value screens candidate feature point again.
S43, on the basis of step S42, the characteristic point of these candidates utilizes extreme value and the zero crossing of principal curvatures Values etc. identify various Geometrical discontinuity point further, finally these characteristic points are extracted from a cloud.
In described step 2, regular geometric body in cloud data is fitted by A25.Extraction in step A24 had limit The cloud data of boundary's feature carries out edge fitting, and judges whether the solid simulated is regular solid.If rule Then solid, then estimate the canonical parameter of regular geometric body;If not regular geometric body, then terminate this step.Concrete bag Containing following steps:
Cuboid matching in S51, cloud data.Cloud data for extracting in step A24 presents the feature of cuboid. The equation of space plane is ax+by+cz=d, (a b c)TRepresent the unit vector in plane normal direction, i.e. a2+b2+c2=1, d Distance for zero to this plane.For three dimensional point cloud, substitute point coordinate (xi yi zi)T, by method of least square, Obtain error equation
Further, according to a2+b2+c2=1 obtains restrictive condition:
(a0, b0, c0, 0) and (δ a, δ b, δ c, δ d)T=0.
Further, calculate parameter value according to the indirect difference of two squares of restrictive condition, simulate six faces respectively, obtain six Parameter (a in facei, bi, ci, di)T
Further, by formulaObtain the apex coordinate of cuboid.
With reference to previous step, obtain the coordinate figure on each summit of cuboid respectively, and estimate each rib long value of cuboid.
Circle matching in S52, cloud data.Cloud data for extracting in step A24 presents round feature.Circle in space Canonical form be (x-a)2+(y-b)2+(z-c)2=r2, it is desirable to solve a circle, at least need four the most point-blank Point.For three dimensional point cloud, choose four data points vi, vj, vk, vm, the normal equation substituting into circle tries to achieve the parameter waiting circle (aijkm, bijkm, cijkm), rijkm
Further, calculating takes other points and distances d of candidate's round edge circle in boundary point cloudl→ijkm.For on border a bit, The distance of candidate's round edge circle and point is
Further, a given distance threshold, it is judged that dl→ijkmWith threshold value TdSize, if dl→ijkm≤Td, then this point exists On candidate's circle.
Further, simulate circle in cloud data, and provide round radius.
S53, simulate regular geometric body present in three dimensional point cloud, and estimate the feature ginseng of regular geometric body Number.
Step 3, build model database, by with the other three dimensional point cloud of marking class circumscribed rectangular body and some cloud in Regular geometric body store hard disk, specifically include:
In described step 3, A31 is for each object to be identified, gathers each object cloud data, obtains marking class other Multiple three dimensional point cloud.
In described step 3, the cloud data gathered is filtered simplifying by A32 respectively.Specifically include following steps:
G21, the cloud data gathered is carried out noise-removed filtering.Read in pending three dimensional point cloud.
Further, set up the topological relation between each point in three dimensional point cloud, set up KD-tree, and be denoted as arbitrfary point piK nearest neighbor be N (p).
Further, current given reference point p is calculatediAnd its k nearest neighbor N (p) average distance each other being denoted as di, can obtain according to range formula:Wherein pjIt is a piK nearest neighbor N (p) in each number Strong point.
Further, threshold value D is set based on experience value, it is judged that the average distance calculated and the magnitude relationship of threshold value, If average distance is less than threshold value, retain this point;If more than threshold value, by this labelling and reject.
Further, travel through all of point, discrete point in final rejecting point cloud, it is achieved denoising.
G22, the cloud data gathered is simplified.Read in three dimensional point cloud, read specified sample percentage.
Further, it is judged that the magnitude relationship between obtained sample percentage and threshold value set in advance, if more than threshold value, Then carry out stochastical sampling only according to threshold value, terminate afterwards;Otherwise carry out stochastical sampling according to the percentage ratio obtained.
Further, set curvature threshold based on experience value, for having carried out the three dimensional point cloud after stochastical sampling, press Carry out curvature sampling according to the threshold value set, and mark the point needing to reject.
Further, the data point of labelling is carried out stochastical sampling, chooses a part of data point, and delete these data points Reject labelling.
Further, labelling is deleted in retrieval again, and deletes these data points, completes to simplify.
In described step 3, A33 asks for the circumscribed rectangular body of each point cloud model and puts the regular geometric body in cloud, builds model Data base.Specifically include following steps:
G31, ask for the circumscribed rectangular body of each point cloud model.
G32, the regular geometric body asked in each point cloud model.
G33, the some cloud feature will asked in step G31 and G32, build model database.
Described step 4 builds model data table, for recognizer quick search.Specifically include following steps:
H11, from given input directory, inquire about all of characteristic model, detect each point cloud feature, utilize Chi- Square distance builds KD-Tree sequence table
H12, K arest neighbors feature of inquiry, and use KDTreeIndex to be indexed.
H13, structure KD-Tree sequence table, be stored to hard disk, for follow-up ONLINE RECOGNITION.
Step 5 carries out ONLINE RECOGNITION to the cloud data inputted, specifically comprises the following steps that
J11, the cloud data inputted is carried out cluster segmentation according to himself feature, obtain treating without classification information The point cloud mass identified.
J12, method according to step 1, for some cloud mass to be identified, ask for taking the circumscribed rectangular body of a cloud, estimate it Rib is long.Obtain the Feature Descriptor to be identified without classification information.
J13, method according to step 2, for some cloud mass to be identified, in estimation cloud data such as cuboid, spheroid The parameters such as on rule solid, estimation cuboid rib is long, radius of sphericity.Obtain the feature description to be identified without classification information Son.
J15, according to constructed model data table and the Feature Descriptor of calculating, quickly inquire about.
Each Feature Descriptor to be identified is retouched by J16, foundation dependency with the feature in constructed model data table State son to mate, it is achieved the identification of cloud data.
It is above the preferably enforcement of the present invention is illustrated, but the invention is not limited to described enforcement Example, those of ordinary skill in the art it may also be made that all equivalent variations on the premise of spirit of the present invention or replacing Changing, deformation or the replacement of these equivalents are all contained in the application claim limited range.

Claims (7)

1. a target identification method based on three dimensional point cloud, it is characterised in that it comprises the following steps:
Three dimensional point cloud other to marking class carries out circumscribed rectangular body to be asked for;
Extracting rule solid from described marking class other three dimensional point cloud;
Build model database;
Build model data table;
Input the cloud data of target to be identified and carry out ONLINE RECOGNITION.
A kind of target identification method based on three dimensional point cloud the most according to claim 1, it is characterised in that described step Rapid three dimensional point cloud other to marking class carries out circumscribed rectangular body to be asked for, and it also includes:
Ask for described some cloud circumscribed rectangular body each crest line direction;
Ask for described some cloud circumscribed rectangular body.
A kind of target identification method based on three dimensional point cloud the most according to claim 1 and 2, it is characterised in that institute State regular geometric body and include cuboid and spheroid.
A kind of target identification method based on three dimensional point cloud the most according to claim 3, it is characterised in that described step Rapid extracting rule solid from described marking class other three dimensional point cloud, also includes:
Use uniform grid method that other some cloud mass of described each marking class is carried out point cloud compressing;
Utilization solves the eigenvalue of sampled point nearest-neighbor point covariance matrix and the method for the normal vector of characteristic vector, to gained To a cloud of simplifying carry out normal vector estimation;
Use curvature method for solving based on local surface, an obtained cloud of simplifying is carried out Curvature Estimate;
Gained is simplified a cloud, according to the some cloud normal vector asked for and curvature, extracts the boundary characteristic of three-dimensional point cloud;
The cloud data with boundary characteristic extracted is carried out edge fitting, and judges that whether the solid simulated is Regular geometric body;The most then estimate the canonical parameter of regular geometric body;Not, then this step is terminated.
A kind of target identification method based on three dimensional point cloud the most according to claim 4, it is characterised in that described step Suddenly model database is built, comprising:
Gather the cloud data of each object to be identified, obtain the other multiple three dimensional point cloud of marking class;
Respectively the cloud data gathered is simplified;
Ask for the circumscribed rectangular body of each point cloud model and put the regular geometric body in cloud, building model database.
A kind of target identification method based on three dimensional point cloud the most according to claim 5, it is characterised in that described step The rapid cloud data inputting target to be identified also carries out ONLINE RECOGNITION, comprising:
The cloud data inputted is carried out cluster segmentation, obtains the some cloud mass to be identified of non-band classification information;
Asking for the circumscribed rectangular body of the cloud data of described to be identified some cloud mass, the feature to be identified obtaining non-band classification information is retouched State son;
Extract the regular geometric body in described cloud data, obtain the Feature Descriptor to be identified of non-band classification information;
According to constructed model data table and the Feature Descriptor of calculating, mate, it is achieved the identification of cloud data.
7. a target identification system based on three dimensional point cloud, it is characterised in that including:
First module, carries out circumscribed rectangular body for execution three dimensional point cloud other to marking class and asks for;
Second module, for performing extracting rule solid from described marking class other three dimensional point cloud;
Three module, is used for performing to build model database;
4th module, is used for performing to build model data table;
5th module, for performing input the cloud data of target to be identified and carry out ONLINE RECOGNITION.
CN201610727765.4A 2016-08-25 2016-08-25 A kind of target identification method based on three dimensional point cloud and system Pending CN106250881A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610727765.4A CN106250881A (en) 2016-08-25 2016-08-25 A kind of target identification method based on three dimensional point cloud and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610727765.4A CN106250881A (en) 2016-08-25 2016-08-25 A kind of target identification method based on three dimensional point cloud and system

Publications (1)

Publication Number Publication Date
CN106250881A true CN106250881A (en) 2016-12-21

Family

ID=57596323

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610727765.4A Pending CN106250881A (en) 2016-08-25 2016-08-25 A kind of target identification method based on three dimensional point cloud and system

Country Status (1)

Country Link
CN (1) CN106250881A (en)

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106951860A (en) * 2017-03-20 2017-07-14 河南腾龙信息工程有限公司 A kind of three-dimensional data intelligent identification Method based on a cloud
CN107247960A (en) * 2017-05-08 2017-10-13 深圳市速腾聚创科技有限公司 Method, object identification method and the automobile of image zooming-out specification area
CN107316048A (en) * 2017-05-03 2017-11-03 深圳市速腾聚创科技有限公司 Point cloud classifications method and device
CN107424185A (en) * 2017-07-27 2017-12-01 深圳前海倍思拓技术有限公司 Conical structure characteristic parameter detecting method based on Point Cloud Processing technology
CN107424189A (en) * 2017-07-27 2017-12-01 深圳前海倍思拓技术有限公司 Ball, cylinder, elliptic cone method of model identification based on Point Cloud Processing technology
CN107832769A (en) * 2017-11-09 2018-03-23 苏州铭冠软件科技有限公司 Object is located at the visual identity method in environment
CN108010002A (en) * 2017-12-25 2018-05-08 浙江大学 A kind of structuring point cloud denoising method based on adaptive implicit Moving Least Squares
CN108133654A (en) * 2018-01-10 2018-06-08 河北农业大学 Cotton plant type design experiment teaching method based on AR mobile phone
CN109145969A (en) * 2018-08-03 2019-01-04 百度在线网络技术(北京)有限公司 Processing method, device, equipment and the medium of three-dimension object point cloud data
CN109344750A (en) * 2018-09-20 2019-02-15 浙江工业大学 A kind of labyrinth three dimensional object recognition methods based on Structural descriptors
CN109581408A (en) * 2018-12-10 2019-04-05 中国电子科技集团公司第十研究所 A kind of method and system carrying out target identification using laser complex imaging
CN109670399A (en) * 2018-11-12 2019-04-23 万瞳(南京)科技有限公司 Museum exhibits image-recognizing method based on computer vision and system
CN109961468A (en) * 2019-03-15 2019-07-02 北京清瞳时代科技有限公司 Volume measuring method, device and storage medium based on binocular vision
CN109993192A (en) * 2018-01-03 2019-07-09 北京京东尚科信息技术有限公司 Recongnition of objects method and device, electronic equipment, storage medium
CN110390346A (en) * 2018-04-23 2019-10-29 北京京东尚科信息技术有限公司 Recongnition of objects method, apparatus, electronic equipment and storage medium
CN110428490A (en) * 2018-04-28 2019-11-08 北京京东尚科信息技术有限公司 The method and apparatus for constructing model
CN110969648A (en) * 2019-12-11 2020-04-07 华中科技大学 3D target tracking method and system based on point cloud sequence data
CN111259958A (en) * 2020-01-15 2020-06-09 北京市商汤科技开发有限公司 Object recognition method and device, and storage medium
CN111707198A (en) * 2020-06-29 2020-09-25 中车青岛四方车辆研究所有限公司 3D vision-based key parameter measurement method for rail vehicle coupler and draft gear
CN115578391A (en) * 2022-12-09 2023-01-06 南京航空航天大学 Feature rapid identification method based on global environment perception
CN117392325A (en) * 2023-11-09 2024-01-12 铁科(北京)轨道装备技术有限公司 Three-dimensional model construction method and device for switch rail

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102750730A (en) * 2012-06-15 2012-10-24 北京理工大学 Characteristic-maintained point cloud data compacting method
CN102890828A (en) * 2012-06-15 2013-01-23 北京理工大学 Point cloud data compacting method based on normal included angle
CN103247041A (en) * 2013-05-16 2013-08-14 北京建筑工程学院 Local sampling-based multi-geometrical characteristic point cloud data splitting method
CN104143210A (en) * 2014-07-31 2014-11-12 哈尔滨工程大学 Multi-scale normal feature point cloud registering method
CN104298971A (en) * 2014-09-28 2015-01-21 北京理工大学 Method for identifying objects in 3D point cloud data

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102750730A (en) * 2012-06-15 2012-10-24 北京理工大学 Characteristic-maintained point cloud data compacting method
CN102890828A (en) * 2012-06-15 2013-01-23 北京理工大学 Point cloud data compacting method based on normal included angle
CN103247041A (en) * 2013-05-16 2013-08-14 北京建筑工程学院 Local sampling-based multi-geometrical characteristic point cloud data splitting method
CN104143210A (en) * 2014-07-31 2014-11-12 哈尔滨工程大学 Multi-scale normal feature point cloud registering method
CN104298971A (en) * 2014-09-28 2015-01-21 北京理工大学 Method for identifying objects in 3D point cloud data

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
卢惠民等: "《ROS与中型组足球机器人》", 31 October 2016, 国防工业出版社 *
唐靖: "三维重建过程中的点云数据处理技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
肖世德著: "《高等机械CAD/CAM》", 31 January 2015, 西南交通大学出版社 *
黄海著: "《虚拟现实技术》", 31 January 2014, 北京邮电大学出版社 *

Cited By (33)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106951860A (en) * 2017-03-20 2017-07-14 河南腾龙信息工程有限公司 A kind of three-dimensional data intelligent identification Method based on a cloud
CN107316048B (en) * 2017-05-03 2020-08-28 深圳市速腾聚创科技有限公司 Point cloud classification method and device
CN107316048A (en) * 2017-05-03 2017-11-03 深圳市速腾聚创科技有限公司 Point cloud classifications method and device
CN107247960A (en) * 2017-05-08 2017-10-13 深圳市速腾聚创科技有限公司 Method, object identification method and the automobile of image zooming-out specification area
CN107424185A (en) * 2017-07-27 2017-12-01 深圳前海倍思拓技术有限公司 Conical structure characteristic parameter detecting method based on Point Cloud Processing technology
CN107424189A (en) * 2017-07-27 2017-12-01 深圳前海倍思拓技术有限公司 Ball, cylinder, elliptic cone method of model identification based on Point Cloud Processing technology
CN107832769A (en) * 2017-11-09 2018-03-23 苏州铭冠软件科技有限公司 Object is located at the visual identity method in environment
CN108010002B (en) * 2017-12-25 2021-06-01 浙江大学 Structured point cloud denoising method based on adaptive implicit moving least square
CN108010002A (en) * 2017-12-25 2018-05-08 浙江大学 A kind of structuring point cloud denoising method based on adaptive implicit Moving Least Squares
CN109993192B (en) * 2018-01-03 2024-07-19 北京京东乾石科技有限公司 Target object identification method and device, electronic equipment and storage medium
CN109993192A (en) * 2018-01-03 2019-07-09 北京京东尚科信息技术有限公司 Recongnition of objects method and device, electronic equipment, storage medium
CN108133654A (en) * 2018-01-10 2018-06-08 河北农业大学 Cotton plant type design experiment teaching method based on AR mobile phone
CN108133654B (en) * 2018-01-10 2019-12-13 河北农业大学 cotton plant type design experiment teaching method based on AR mobile phone
CN110390346A (en) * 2018-04-23 2019-10-29 北京京东尚科信息技术有限公司 Recongnition of objects method, apparatus, electronic equipment and storage medium
CN110428490B (en) * 2018-04-28 2024-01-12 北京京东尚科信息技术有限公司 Method and device for constructing model
CN110428490A (en) * 2018-04-28 2019-11-08 北京京东尚科信息技术有限公司 The method and apparatus for constructing model
CN109145969A (en) * 2018-08-03 2019-01-04 百度在线网络技术(北京)有限公司 Processing method, device, equipment and the medium of three-dimension object point cloud data
CN109145969B (en) * 2018-08-03 2020-07-28 百度在线网络技术(北京)有限公司 Method, device, equipment and medium for processing point cloud data of three-dimensional object
CN109344750A (en) * 2018-09-20 2019-02-15 浙江工业大学 A kind of labyrinth three dimensional object recognition methods based on Structural descriptors
CN109344750B (en) * 2018-09-20 2021-10-22 浙江工业大学 Complex structure three-dimensional object identification method based on structure descriptor
CN109670399A (en) * 2018-11-12 2019-04-23 万瞳(南京)科技有限公司 Museum exhibits image-recognizing method based on computer vision and system
CN109581408A (en) * 2018-12-10 2019-04-05 中国电子科技集团公司第十研究所 A kind of method and system carrying out target identification using laser complex imaging
CN109581408B (en) * 2018-12-10 2023-01-06 中国电子科技集团公司第十一研究所 Method and system for identifying target by using laser composite imaging
CN109961468B (en) * 2019-03-15 2021-08-13 北京清瞳时代科技有限公司 Volume measurement method and device based on binocular vision and storage medium
CN109961468A (en) * 2019-03-15 2019-07-02 北京清瞳时代科技有限公司 Volume measuring method, device and storage medium based on binocular vision
CN110969648B (en) * 2019-12-11 2022-05-20 华中科技大学 3D target tracking method and system based on point cloud sequence data
CN110969648A (en) * 2019-12-11 2020-04-07 华中科技大学 3D target tracking method and system based on point cloud sequence data
CN111259958B (en) * 2020-01-15 2024-04-16 北京市商汤科技开发有限公司 Object recognition method and device and storage medium
CN111259958A (en) * 2020-01-15 2020-06-09 北京市商汤科技开发有限公司 Object recognition method and device, and storage medium
CN111707198A (en) * 2020-06-29 2020-09-25 中车青岛四方车辆研究所有限公司 3D vision-based key parameter measurement method for rail vehicle coupler and draft gear
CN111707198B (en) * 2020-06-29 2021-08-03 中车青岛四方车辆研究所有限公司 3D vision-based key parameter measurement method for rail vehicle coupler and draft gear
CN115578391A (en) * 2022-12-09 2023-01-06 南京航空航天大学 Feature rapid identification method based on global environment perception
CN117392325A (en) * 2023-11-09 2024-01-12 铁科(北京)轨道装备技术有限公司 Three-dimensional model construction method and device for switch rail

Similar Documents

Publication Publication Date Title
CN106250881A (en) A kind of target identification method based on three dimensional point cloud and system
WO2024077812A1 (en) Single building three-dimensional reconstruction method based on point cloud semantic segmentation and structure fitting
CN109410321B (en) Three-dimensional reconstruction method based on convolutional neural network
Guo et al. 3D object recognition in cluttered scenes with local surface features: A survey
Aldoma et al. Tutorial: Point cloud library: Three-dimensional object recognition and 6 dof pose estimation
CN108376408B (en) Three-dimensional point cloud data rapid weighting registration method based on curvature features
CN112101430B (en) Anchor frame generation method for image target detection processing and lightweight target detection method
CN103488972B (en) Fingertip Detection based on depth information
CN114972459B (en) Point cloud registration method based on low-dimensional point cloud local feature descriptor
Mozos et al. Furniture models learned from the www
CN106204557A (en) A kind of extracting method of the non-complete data symmetrical feature estimated with M based on extension Gaussian sphere
CN110111375B (en) Image matching gross error elimination method and device under Delaunay triangulation network constraint
CN115861397A (en) Point cloud registration method based on improved FPFH-ICP
CN107025449A (en) A kind of inclination image linear feature matching process of unchanged view angle regional area constraint
Dickinson et al. Integrating qualitative and quantitative shape recovery
CN113706381A (en) Three-dimensional point cloud data splicing method and device
CN104851095A (en) Workpiece image sparse stereo matching method based on improved-type shape context
CN114119922A (en) Complex contour defect checking method based on point cloud three-dimensional reconstruction
CN110334704B (en) Three-dimensional model interest point extraction method and system based on layered learning
Gong et al. A two-level framework for place recognition with 3D LiDAR based on spatial relation graph
Liao et al. Fuzzypsreg: Strategies of fuzzy cluster-based point set registration
CN109598261B (en) Three-dimensional face recognition method based on region segmentation
CN116091727A (en) Complex Qu Miandian cloud registration method based on multi-scale feature description, electronic equipment and storage medium
CN117928385A (en) Engineering construction intelligent measurement method based on remote unmanned aerial vehicle and sensor
Qiao et al. G3reg: Pyramid graph-based global registration using gaussian ellipsoid model

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20161221