CN108665472A

CN108665472A - The method and apparatus of point cloud segmentation

Info

Publication number: CN108665472A
Application number: CN201710214706.1A
Authority: CN
Inventors: 胡事民; 杨晟; 陈康; 张维; 刘健庄
Original assignee: Tsinghua University; Huawei Technologies Co Ltd
Current assignee: Tsinghua University; Huawei Technologies Co Ltd
Priority date: 2017-04-01
Filing date: 2017-04-01
Publication date: 2018-10-16

Abstract

This application provides a kind of methods and apparatus of point cloud segmentation, during point cloud segmentation, aspect is determined by interactive mode, and under aspect by pending point cloud segmentation be several complete parts of semanteme, complete label and segmentation in real time, splitting speed is substantially improved, and reduces user annotation amount.This method includes：The point treated in process points cloud merges, and obtains multiple point sets；Visual angle sampling is carried out under a multiple of viewing angles to the pending cloud, obtains the view under each visual angle in multiple visual angle；View under each visual angle in multiple visual angle is assessed, and aspect is determined from multiple visual angle according to assessment result；After determining the aspect, the image of the view instruction under the aspect is shown；Obtain the seed point set of the mark object of user's mark；According to the seed point set, the corresponding point set of mark object is concentrated to be split with other point sets multiple point.

Description

The method and apparatus of point cloud segmentation

Technical field

This application involves image processing fields, and more particularly, to a kind of method and apparatus of point cloud segmentation.

Background technology

With the maturation of virtual reality technology and popularizing for three-dimensional scanning device, acquisition and the processing for putting cloud are just further square Just.During the treatment, a most basic link, that is, point cloud segmentation, i.e., by the collected point cloud segmentation of institute at multiple Complete semantic part, the process are divided into automatic segmentation and two class of Interactive Segmentation according to whether needing human intervention.However, existing The result controllability of some automatic segmentation algorithms is not high, often there is over-segmentation (Over-Segmentation), less divided (Under-Segmentation), undivided (Un-segmented) and wrong segmentation, existing Interactive Segmentation algorithm, which exists, to be handed over Mutually measure big disadvantage.

Therefore, how segmentation accuracy is improved during point cloud segmentation and reduce interactive quantity, be one urgently to be resolved hurrily The problem of.

Invention content

The application provides a kind of method and apparatus of point cloud segmentation, can pass through interactive mode during point cloud segmentation It determines aspect, and is several complete parts of semanteme by pending point cloud segmentation under aspect, complete mark in real time Note and segmentation are substantially improved splitting speed, and reduce the mark amount of user.

In a first aspect, a kind of method of point cloud segmentation is provided, including：The point treated in process points cloud merges, and obtains To multiple point sets；Visual angle sampling is carried out under a multiple of viewing angles to the pending cloud, is obtained in multiple visual angle under each visual angle View, multiple visual angle is the visual angle of at least two representative scenes in the pending cloud；To in multiple visual angle View under each visual angle is assessed, and aspect is determined from multiple visual angle according to assessment result；It should determining After aspect, the image of the view instruction under the aspect is shown；Obtain the seed point set of user's mark, the seed point Collection is to mark the corresponding part point set of object in the view under the aspect in multiple point set；It is right according to the seed point set Multiple point concentrates the corresponding point set of mark object to be split with other point sets.

Therefore, in the method for the point cloud segmentation of the embodiment of the present application, aspect is determined during point cloud segmentation, The seed point set of user's mark is obtained by interactive mode under aspect, and according to seed point set pair point cloud minute It cuts, reduces interactive quantity of the user on selection visual angle, while reducing mark amount of user during determining seed point set, Improve point cloud segmentation speed.

Optionally, in a kind of realization method of first aspect, this is treated the point in process points cloud and merges, and obtains more A point set, including：

The normal direction of the curvature and the consecutive points each put put according to each of the pending cloud, normal direction projection Distance and color space at least one of are estimated apart from equidistant, and the point of the pending cloud is merged into multiple point set.

Optionally, in a kind of realization method of first aspect, the curvature put according to each of the pending cloud, And normal direction, normal direction projector distance and the color space of the consecutive points each put at least one of are estimated apart from equidistant, The point of the pending cloud is merged into multiple point set, including：

Choose the point that multiple point set and curvature minimum are not yet merged into the pending cloud；

Determine that this not yet merges into the normal direction of multiple point set and the consecutive points of the point of curvature minimum, normal direction projector distance At least one of estimate apart from equidistant with color space, the consecutive points be not yet merge into multiple point set point and/or The center of mass point for the point set that multiple point is concentrated；

When the normal direction of the consecutive points, normal direction projector distance and color space apart from it is equidistant at least one of estimate be less than When first threshold, the point that this is not yet merged into multiple point set and curvature minimum merges into multiple point set with the consecutive points In a point set.

Choose the point that the first point set and curvature minimum are not yet merged into the pending cloud；

Determine that this not yet merges into the normal direction of first point set and the consecutive points of the point of curvature minimum, normal direction projector distance At least one of estimate apart from equidistant with color space, the consecutive points be not yet merge into first point set point and/or The center of mass point of first point set；

When the normal direction of the consecutive points, normal direction projector distance and color space apart from it is equidistant at least one of estimate be less than When first threshold, this is not yet merged into the point of first point set and curvature minimum and the consecutive points merge at least two the One point set；

This at least two first points concentrations are chosen not yet to merge into multiple point set and include first points of points at most Collection；

Determine this not yet merge into multiple point set and include the adjacent point set of the first most point set of points fitting it is flat At least one of normal direction, barycenter and the color histogram in face, the adjacent point set are not yet to merge into the first of multiple point set The point set that point set and/or multiple point are concentrated；

When at least one of the normal direction of the fit Plane of the adjacent point set, barycenter and color histogram are less than second threshold When, this is not yet merged into multiple point set and includes most the first point set point set adjacent with this of points and is multiple The point set that point is concentrated；

When at least one of the normal direction of the fit Plane of the adjacent point set, barycenter and color histogram are more than or equal to When second threshold, determine that this is not yet merged into multiple point set and includes that the first most point set of points is concentrated for multiple point A point set.

Therefore, in the method for the point cloud segmentation of the embodiment of the present application, during point cloud segmentation, the point in cloud is led to Cross repeatedly to merge and obtain final point set, at the same introduce during point set merges the fit Plane of adjacent point set normal direction, The factors such as barycenter and color histogram improve the accuracy of point set merging, reduce problem scale, reach practicability requirement.

Optionally, in a kind of realization method of first aspect, this method further includes：

The number for the first point set that can not merge into multiple point set is concentrated to be more than third threshold when this at least two first points When value, the second threshold is adjusted at least once, so that the first point set that can not merge into multiple point set partly or entirely merges For multiple point set.

Therefore, in the method for the point cloud segmentation of the embodiment of the present application, during point cloud segmentation, when pending cloud When point cloud is second-rate, point set can also be merged, that is reduced by way of adjusting the threshold value of point set merging at least once The quantity of a little only point sets comprising base point, preferable point cloud segmentation effect can also be reached when cloud is ropy by realizing Fruit reaches practicability requirement.

Optionally, in a kind of realization method of first aspect, this concentrates the mark according to the seed point set, to multiple point The corresponding point set of note object is split with other point sets, including：

According to the seed point set, determine that multiple point concentrates the corresponding pending point set of the mark object, the pending point Collection is that the corresponding point of the mark object concentrates the arbitrary point set in addition to the seed point set；

According to the pending point set, the corresponding point set of mark object and other point sets minute are concentrated to multiple point It cuts.

Optionally, in a kind of realization method of first aspect, according to the pending point set, being concentrated to multiple point should for this The corresponding point set of mark object is split with other point sets, including：

The area of convex closure is corresponded to according to multiple point concentration level point collection, vertical point collection corresponds to the area of convex closure, Yi Jishui Spatial neighborhood relations between flat spot collection and vertical point set, between determining that multiple point concentrates two point sets for constituting point set pair Supporting relation, wherein the horizontal point set is the point set that multiple point concentration is less than 45 ° with ground angle, which is should Multiple points concentrate the point set with ground angle more than or equal to 45 °, a horizontal point set and a vertical point set to constitute one Point set pair；

The supporting relation between two point sets for constituting point set pair is concentrated according to multiple point, and it is every to determine that multiple point is concentrated Spatial neighborhood relations between a point set；

The spatial neighborhood relations between each point set are concentrated according to multiple point, determine the pending point set and the seed point Mapping relations between collection；

It is right according to the pending point set after determining the mapping relations between the pending point set and the seed point set Multiple point concentrates the corresponding point set of mark object to be split with other point sets.

Optionally, in a kind of realization method of first aspect, this concentrates level point collection to correspond to convex closure according to multiple point Area, vertical point collection correspond to the spatial neighborhood relations between the area and horizontal point set and vertical point set of convex closure, determining should Multiple points concentrate the supporting relation between two point sets for constituting point set pair, including：

According to formulaDetermine that multiple point concentrates vertical point collection p_iWith horizontal point set p_jThe vertical point set p of point set centering of composition_iWith horizontal point set p_jBetween supporting relation,

Wherein, Q<p_i,p_j>Indicate the vertical point set p of composition point set pair_iWith horizontal point set p_jBetween supporting relation, p_iIt is I-th of vertical point set, p_jIt is j-th of horizontal point set, W_iIt is point set p_iThe area of corresponding convex closure, W_jIt is point set p_jThe face of corresponding convex closure Product, α are predefined parameter, U (p_i,p_j) be and p_iIt is adjacent and equally support p_jVertical point set number, as U (p_i,p_j) ≠ 0 is Vertical point set p_iSupport level point set p_j, U (p_i,p_j)=0 is horizontal point set p_jSupport vertical point set p_i。

Optionally, in a kind of realization method of first aspect, this concentrates the sky between each point set according to multiple point Between neighbouring relations, determine the mapping relations between the pending point set and the seed point set, including：

According to the data cost and the pending point set and the pending point between the seed point set and pending point set Smooth cost between the adjacent point set of collection, determines the mapping relations between the pending point set and the seed point set,

Wherein, which is difference by the plane characteristic between pending point set and seed point set, pending point Space length, pending point set between the center of collection and the center of nearest seed point set and the color histogram between seed point set The difference of figure determines；The smooth cost is generated when adjacent point set belongs to a different category with pending point set by pending point set The cost of color characteristic, the fit Plane that generates when belonging to a different category of adjacent point set of pending point set and pending point set The supporting relation feature that the adjacent point set of the cost of feature, pending point set and pending point set generates when belonging to a different category Cost determines.

Optionally, in a kind of realization method of first aspect, this is according between the seed point set and pending point set Smooth cost between data cost and the pending point set and the adjacent point set of the pending point set, determines that this is pending Mapping relations between point set and the seed point set, including：

According to formulaDetermine the pending point set and the seed point set Between mapping relations,

Wherein, E indicates the mapping relations between the pending point set and the seed point set, E_D(p_i,v_i)=λ_DD_D(p_i, U_i)+λ_CD_C(p_i,U_i)+λ_PD_P(p_i,U_i)+λ_SD_S(p_i,U_i), data cost E_DIt is pending point set p_iIt is classified as label v_iGeneration Valence function, p_iIt is pending point set, v_iIt is the corresponding label of seed point collection； E_S(p_i,v_i,p_j,v_j)=λ_C′S_C(p_i,v_i,p_j,v_j) +λ_P′S_P(p_i,v_i,p_j,v_j)+λ_S′S_S(p_i,v_i,p_j,v_j), the smooth cost E_SIt is the phase of pending point set and pending point set Point set pair (the p of adjoint point collection composition_i,p_j) it is classified as label (v respectively_i,v_j) cost function, p_jIt is the consecutive points of pending point set Collection, v_jIt is the corresponding label of consecutive points collection of seed point set；U_iIt is to be classified as label v_iAll seed point sets constitute set； D_DIt is the cost of distance feature, depends on p_iCenter and the center of nearest seed point set between space length；D_CIt is color characteristic Cost, depend on p_iWith the difference of the color histogram of seed point set；D_PIt is the cost of plane characteristic, depends on p_iAnd seed The difference of plane characteristic (distance of normal direction and normal direction) between point set；D_SIt is the cost of supporting relation feature, depends on p_iAnd seed Supporting relation between point set；S_CIt is to work as p_iAnd p_jLabel (the v to belong to a different category_i≠v_j) when generation of color characteristic for generating Valence；S_PIt is to work as p_iAnd p_jLabel (the v to belong to a different category_i≠v_j) when the cost of fit Plane feature that generates；S_SIt is to work as p_iAnd p_j Label (the v to belong to a different category_i≠v_j) when the cost of supporting relation feature that generates；λ is constant, λ_D、λ_C、λ_S、λ_C′、λ_P′ And λ_S' constant.

Therefore, during the point cloud segmentation of the embodiment of the present application, pass through the pending point set of above-mentioned determination and seed point set Between mapping relations function, determine the mapping relations between pending point set and seed point set, in turn, treating process points When the corresponding point set of mark object is split with other point sets in cloud, the speed and accuracy of segmentation are improved.

Optionally, in a kind of realization method of first aspect, this to the view under each visual angle in multiple visual angle into Row assessment, and aspect is determined from multiple visual angle according to assessment result, including：

Show the image of the view instruction in multiple visual angle under each visual angle；

Obtain marking result of the user to the view under each visual angle in multiple visual angle；

According to the marking as a result, determining the priority at multiple visual angle；

Determine that the visual angle of highest priority is the aspect.

Extract the projected area A of the point in the view in multiple visual angle under each visual angle_pa, point comentropy A_vaAnd point Density A_pd；

According to the projected area A of the point in the view under each visual angle_pa, point comentropy A_vaWith the density A of point_pd, really The fixed aspect.

Optionally, in a kind of realization method of first aspect, the throwing according to the point in the view under each visual angle Shadow area A_pa, point comentropy A_vaWith the density A of point_pd, determine the aspect, including：

According to the marking as a result, determining the projected area A of the point in the view in multiple visual angle under each visual angle_pa, point Comentropy A_vaWith the density A of point_pdWeight；

According to formulaDetermine the aspect in multiple visual angle；

Wherein, V_n+1Indicate the aspect, Y_i=β^T·A_i, p_iIt is all point sets under the i of visual angle, p_nIt is the institute under the n of visual angle Pointed set, visual angle n are a upper visual angle of visual angle i, and β is the projected area A put under the i of visual angle_pa, point comentropy A_vaWith point Density A_pdWeight composition three-dimensional vector, A_iIt is the projected area A put under the i of visual angle_pa, point comentropy A_vaWith the density of point A_pdThe three-dimensional vector of composition.

During point set is divided, the letter of the point in the view in current multiple visual angle under each visual angle is calculated in real time Cease entropy A_va', and reacquire again marking result of the user to the view under each visual angle in current multiple visual angles；

According to the comentropy A of the point in the view under each visual angle calculated in real time_va' and this give a mark again as a result, really The projected area A of the point in view in fixed current multiple visual angles under each visual angle_pa' density the A with point_pd′；

According to formulaDetermine the aspect in current multiple visual angles；

Wherein, V_n+1Indicate the aspect, Y_i=β^T·A_i', p_iIt is all point sets under the i of visual angle, p_nIt is under the n of visual angle All point sets, visual angle n are a upper visual angle of visual angle i, and β is the projected area A put under the i of visual angle_pa', point comentropy A_va' and The density A of point_pd' weight composition three-dimensional vector, A_i' it is the projected area A put under the i of visual angle_pa', point comentropy A_va' and The density A of point_pdThe three-dimensional vector of ' composition.

Therefore, during the point cloud segmentation of the embodiment of the present application, aspect is determined by interactive mode, and in target Point cloud segmentation is carried out under visual angle, reduces the interactive quantity that user selects visual angle, while hoist point cloud splitting speed.

Second aspect provides a kind of equipment of point cloud segmentation, can execute first aspect or any of first aspect can The module or unit of method in the realization method of choosing.

The third aspect provides a kind of equipment of point cloud segmentation, including memory and processor, is stored on the memory The program code for executing above-mentioned first aspect or its arbitrary optional realization method is can serve to indicate that, when the code is performed When, which may be implemented each operation that equipment executes in the method for first aspect.

Fourth aspect provides a kind of computer storage media, and have program stored therein code in the computer storage media, should Program code can serve to indicate that the method in the arbitrary optional realization method for executing above-mentioned first aspect or first aspect.

5th aspect, it includes the computer program product instructed to provide a kind of, when run on a computer so that Computer executes the method described in above-mentioned various aspects.

Description of the drawings

Fig. 1 is the schematic block diagram according to the three-dimensional scanning device of the embodiment of the present application.

Fig. 2 is the schematic flow chart according to the method for the point cloud segmentation of the embodiment of the present application.

Fig. 3 is the schematic diagram according to the flow of the point cloud segmentation of the embodiment of the present application.

Fig. 4 is the example according to a point cloud segmentation of the embodiment of the present application.

Fig. 5 is the example according to another point cloud segmentation of the embodiment of the present application.

Fig. 6 is the example according to another point cloud segmentation of the embodiment of the present application.

Fig. 7 is the example according to another point cloud segmentation of the embodiment of the present application.

Fig. 8 is the schematic block diagram according to the equipment of the point cloud segmentation of the embodiment of the present application.

Fig. 9 is the schematic block diagram according to the equipment of the point cloud segmentation of the embodiment of the present application.

Specific implementation mode

Below in conjunction with attached drawing, the technical solution in the application is described.

It should be understood that in the embodiment of the present application, the equipment of point cloud segmentation can be some setting with 3-D scanning function It is standby, for example, laser radar apparatus (Lidar Devices), Microsoft Kinect, master reference (Prime Sensor), three-dimensional knot Structure scanner (Structure Sensor) etc..

Fig. 1 is the schematic block diagram using a kind of three-dimensional scanning device 100 of the method for point cloud segmentation of the application.Such as figure Shown in 1, which includes：Camera 110, processor 120, memory 130, input unit 140 and display Unit 150.It will be understood by those skilled in the art that the structure of three-dimensional scanning device 100 shown in Fig. 1 is not constituted to three-dimensional The restriction of scanning device may include either combining certain components or different components than illustrating more or fewer components Arrangement.

Camera 110 can be used for shooting 3-D view, and/or can be with sampling depth data.

Processor 120 is the control centre of three-dimensional scanning device 100, is swept using various interfaces and the entire three-dimensional of connection The various pieces for retouching equipment 100, by running or executing the software program and/or module that are stored in memory 130, and tune With the data being stored in memory 130, the various functions and processing data of three-dimensional scanning device 100 are executed, to three-dimensional Scanning device 100 carries out integral monitoring.Optionally, processor 120 may include one or more processing units；Preferably, it handles Device 120 can integrate application processor, the main processing operation system of application processor, user interface and application program etc..

Memory 130 can be used for storing software program and module, and processor 120 is stored in memory 130 by operation Software program and module, to execute various function application and the data processing of three-dimensional scanning device 100.Memory 130 can include mainly storing program area and storage data field, wherein storing program area can storage program area, at least one work( Application program (such as shooting function, point cloud segmentation function, visual angle recommendation function etc.) needed for energy etc.；Storage data field can store Created data (such as photographed data, point cloud segmentation data, visual angle recommending data are used according to three-dimensional scanning device 100 Deng) etc..In addition, memory 130 may include high-speed random access memory, can also include nonvolatile memory, such as At least one disk memory, flush memory device or other volatile solid-state parts.

Input unit 140 can be used for receiving the number or character information of input, and generate and three-dimensional scanning device 100 User setting and the related key signals input of function control.Specifically, which can include but is not limited to touch It is one or more in screen, physical keyboard, trace ball, mouse, operating lever etc..

Display unit 150 can be used for showing information input by user or the information of user and 3-D scanning be supplied to set Standby 100 various menus.

Optionally, which can be used cooperatively with terminal devices such as computer, mobile phones.

Fig. 2 is the schematic flow chart according to a kind of method 200 of point cloud segmentation of the embodiment of the present application.This method 200 It can be executed by some three-dimensional scanning devices with point cloud segmentation function, for example, can be executed by three-dimensional structure scanner, when So, it can also be executed, can also be executed with three-dimensional scanning device and computer, the application is real by other three-dimensional scanning devices Example is applied to be particularly limited not to this.This method 200 includes：

In 210, the point treated in process points cloud merges, and obtains multiple point sets.

It is alternatively possible to the method for the curvature and the consecutive points each put put according to each of the pending cloud At least one of estimate apart from equidistant to, normal direction projector distance and color space, the point of the pending cloud is merged into Multiple point set.

It should be understood that in the pending cloud, each point has spatial position coordinate and colouring information.

Optionally, which can be three-dimensional point cloud.

Optionally, it in 210, converts the base unit of point cloud segmentation to point set by putting, to reduce problem scale, reaches To practicability requirement.

Specifically, multiple point set can be obtained by two ways.

Mode one：

A single point in the pending cloud is merged into point set it should be understood that being realized in mode one.

Specifically, structure three-dimensional point cloud atlas G₀=<V₀,E₀>, wherein V₀Indicate point set, V₀It include the institute in three-dimensional point cloud A little, E₀Indicate side collection, E₀Include choosing V first in each round merging process by all side collection that k nearest neighbor obtains₀ In be not yet combined into multiple point set and the point of curvature minimum, its consecutive points is considered one by one, when the normal direction of consecutive points, normal direction project Distance and color space apart from it is equidistant at least one of estimate be less than first threshold when, this is not yet merged into multiple point Collection and the point of curvature minimum merge into a point set of multiple point concentration with the consecutive points, at this point, the matter of the point set after merging Heart point can also be used as the consecutive points of other point, in V₀In all the points when belonging to a certain point set, end point merges into point set Flow.

It should be understood that having merged into point set with those when this not yet merges into multiple point set and the point of curvature minimum When center of mass point merges, its essence is the points that the point for not yet merging into multiple point set and curvature minimum has merged with those Collection merges.

Optionally, which is a changeable threshold value, can need to set according to practical merging.

Optionally, consecutive points can be the point that the pending point Yun Zhongwei merges into point set, can also be those economic cooperation And for point set center of mass point, can also exist simultaneously the pending point Yun Zhongwei merge into point set point and those merged into The center of mass point of point set.

It is alternatively possible in the presence of containing only, there are one the point sets of point.

Mode two：

It should be understood that in mode two, the first step merges：A single point in the pending cloud is merged at least two first Point set, second step merge：At least two first points of some or all of of concentration are merged into multiple point set.

It should also be understood that the point set that first point set generates after merging for the first step, the first step merges in second step merging The first point set generated afterwards partly or entirely merges into multiple point set, and the first point set not merged can also be used as multiple point The part of concentration.

Specifically, the first step merges：Build view G₀=<V₀,E₀>, wherein V₀Indicate point set, V₀It include three-dimensional point cloud In all the points, E₀Indicate side collection, E₀Include all side collection obtained by k nearest neighbor, in each round merging process, first Choose V₀In be not yet combined into first point set and the point of curvature minimum, consider its consecutive points one by one, normal direction, method when consecutive points To projector distance and color space apart from it is equidistant at least one of estimate be less than first threshold when, this is not yet merged into the One point set and the point of curvature minimum merge into a point set of this first point concentration with the consecutive points, at this point, the point set after merging Center of mass point can also be used as the consecutive points of other point, in V₀In all the points when belonging to a certain first point set, end point is closed And it is the flow of the first point set.Second step merges：Build view G₁=<V₁,E₁>, wherein V₁Indicate the first point set, V₁Include All first point sets, E₁Indicate side collection, E₁Include passing through G₀Syntople formed all side collection, merged in each round Cheng Zhong chooses V first₁In be not yet combined into multiple point set and include the first most point set of points, consider that its is adjacent one by one Point set considers the essential attributes such as normal direction, barycenter and the color histogram of fit Plane of each first point set, when the adjacent point set At least one of normal direction, barycenter and the color histogram of fit Plane when being less than second threshold, this is not yet merged into this Multiple point sets and include most the first point set point set adjacent with this of points and for a point set of multiple point concentration, this When, the point set for merging into multiple point set can also be used as the adjacent point set of other first point set, in V₁In all first points Collect after being involved in second step merging, the part that the first point set that those do not merge is also concentrated as multiple point.

Optionally, the first threshold and the second threshold are changeable threshold value, can need to set according to practical merging.

It is alternatively possible in the presence of containing only, there are one the first point sets of point.

Optionally, adjacent point set can be the first point set for not yet merging into multiple point set, can also be those The point set that multiple point is concentrated is merged into, can also exist simultaneously the first point set for not yet merging into multiple point set and that A little point sets merged into multiple point and concentrated.

Optionally, color histogram can be HSV (Hue, Saturation, Value) color histogram, can also be RGB (Red, Green, Blue) color histogram.

Optionally, the point cloud quality of the pending cloud is poorer, this at least two first points concentrations merge into multiple point The number of first point set of collection is fewer.

It should be understood that point cloud it is of poor quality can be the pending cloud point cloud Density inhomogeneity, can also be that this waits locating There is cavity in reason point cloud, can also be that point cloud caused by some other factors is of poor quality.

Optionally, when this at least two first points are concentrated the number for the first point set that can not merge into multiple point set to be more than When third threshold value, the second threshold is adjusted at least once, so as to which the first point set part or complete of multiple point set can not be merged into Portion merges into multiple point set.

Optionally, which is changeable threshold value, can need to set according to practical merging.

Optionally, the number for the first point set that can not merge into multiple point set is concentrated to be more than the at this at least two first points Three threshold values can be caused by cloud is of poor quality.

Optionally, adjustment second threshold refers to reducing second threshold to merge into this at least two first point set more every time The limitation of a point set every time suitably reduces threshold value, to reduce the quantity of scrappy point set.

It should be understood that every time adjustment second threshold when, all the first point sets for not merging into multiple point set all need again into The following operation of row：

When at least one of the normal direction of the fit Plane of the adjacent point set, barycenter and color histogram are less than second threshold When, this is not yet merged into multiple point set and includes most the first point set point set adjacent with this of points and is multiple The point set that point is concentrated.

It should also be understood that there are two criterion for the extraction of point set：One, merge as possible belong to the same semantic object o'clock to one Point set；Two, should not occur include the point of different semantic objects point set.

In 220, visual angle sampling is carried out under a multiple of viewing angles to the pending cloud, obtains each regarding in multiple visual angle View under angle, multiple visual angle are the visual angles of at least two representative scenes in the pending cloud.

It should be understood that the quantity at the visual angle of the pending cloud is very huge, it, can be tight if handled each visual angle Ghost image rings point cloud segmentation process, while amount of user interaction is also very big, therefore, the visual angle of representative scene is selected to be regarded Angle samples.

It should also be understood that multiple visual angle be the pending cloud visual angle in minority can clearly reflect the pending point The visual angle of representative scene in cloud.

It is alternatively possible to multiple visual angle is chosen in all visual angles of the pending cloud by interactive mode, Multiple visual angle can only be generated.

Optionally, visual angle can be obtained by following flow：

Pending cloud scene is placed in the coordinate Manhattan Coordinate of Manhattan first, and makes Z axis just It is directed toward upwardly direction perpendicular to the ground in top.Then, using traditional stochastical sampling consistency algorithm (Random Sample The methods of Consensus, RANSAC), horizontal plane is extracted, floor segmentation is carried out.If there are ceilings for scene, by it As highest first floor.

After extracting floor, each layer is sampled (View Sampling).The viewpoint of required sampling is divided into two classes：It bows Depending on visual angle (Top-Down Views) and conventional visual angle (Interior Perspectives).Wherein, overlook visual angle with (x, y, Z, φ) it indicates, conventional visual angle indicates that φ is subtended angle, θ is mobile visual angle with (x, y, z, θ, φ).Overlooking visual angle can be by every floor 2D bounding boxs determine, can obtain varigrained vertical view in different height acquisition, it is more Gao Yueguang, about low about thin；It is conventional For viewpoint by carrying out uniform sampling acquisition inside scene, conventional visual angle carries out sampling on the basis of conventional viewpoint to θ, φ can .

It should be understood that in the coordinate Manhattan Coordinate of Manhattan, it may be determined that X-axis, Y-axis and Z axis.

In 230, the view under each visual angle in multiple visual angle is assessed, and more from this according to assessment result Aspect is determined in a visual angle.

It is alternatively possible to be assessed the view under each visual angle in multiple visual angle by user.

Optionally, which can be the highest visual angle of user's scoring, i.e. optimal viewing angle.

Specifically, the aspect can be determined as follows.

Mode one：

Determine that the visual angle of highest priority is the aspect.

Optionally, after the image of the view instruction in showing multiple visual angle under each visual angle, user can be direct Obtain in multiple visual angle under each visual angle view instruction image, and under each visual angle view instruction image into Row marking.

Optionally, user can give a mark to the view under each visual angle in multiple visual angle respectively, and scoring mechanism can be with Be 0-5, can also be 0-100, can also be the marking of some other forms, such as it is excellent, good, neutralize difference, the application is simultaneously unlimited It is formed on this.User judges the view under each visual angle in multiple visual angle according to itself, is each in multiple visual angle View under visual angle is given a mark.

Optionally, it gives a mark higher, priority is higher.

Mode two：

The aspect in multiple visual angle is determined according to following formula 1,

Mode three：

The aspect in current multiple visual angles is determined according to following formula 2,

It should be understood that in mode two and mode three, user's marking mode is similar to the mode of mode one, for sake of simplicity, This is repeated no more.

Optionally, in mode two and mode three, the descriptor put under multiple visual angle includes projected area A_pa, comentropy A_vaWith density A_pd, the descriptor put under multiple visual angle can also include some other factors, and the application be not restricted to This.

It optionally, can be according to the marking at multiple visual angle as a result, determining multiple visual angle in mode two and mode three In the projected area A that puts under each visual angle_pa, point comentropy A_vaWith the density A of point_pdWeight.For example, by each visual angle Marking achievement is averagely allocated to projected area A a little_pa, point comentropy A_vaWith the density A of point_pd, and then determine the perspective plane of point Product A_pa, point comentropy A_vaWith the density A of point_pdWeight.

Optionally, higher, the projected area A of the point under corresponding visual angle of user's marking_paWeight, point comentropy A_va's The density A of weight and point_pdWeight it is higher.

Optionally, in mode two and mode three, the comentropy put under each visual angle in multiple visual angle is divided in point set During, it can change in real time.The point set not marked is more, and comentropy is bigger, i.e., during point set is divided, letter Breath entropy can become smaller and smaller.

In 240, after determining the aspect, the image of the view instruction under the aspect is shown.

Optionally, after determining the aspect, user can obtain the image of the instruction of the view under aspect.

In 250, the seed point set of user's mark is obtained, which is multiple in the view under the aspect The corresponding part point set of object is marked in point set.

Optionally, after the image that user obtains the view instruction under aspect, user is under the aspect The corresponding part point set of mark object is chosen in the image of view instruction as seed point set, for example, choosing the mark object corresponding 3 A point set is as seed point set, at this point, the mark object can correspond to 20 point sets.

It should be understood that the image of the view instruction under the aspect includes at least one mark object, user for this at least Object, which is each marked, in one mark object chooses corresponding seed point set.

It should also be understood that user, which when choosing the seed point set, can be the seed point set, assigns the mark distinguished with other point sets Label.

It should also be understood that the view under the aspect is a two dimension view.

Optionally, user chooses the corresponding part point set of the mark object as seed point set, i.e. seed point set can be A few point set that user chooses.

In 260, according to the seed point set, to multiple point concentrate the corresponding point set of mark object and other point sets into Row segmentation.

It should be understood that concentrate the corresponding point set of mark object to be split with other point sets multiple point, its essence is Assign corresponding with mark object seed point set identical label for the corresponding all point sets of the mark object, that is, by the mark object Corresponding point set concentrates other point sets to distinguish with multiple point.

It optionally, can be according to image segmentation Graph-Cut algorithms to multiple point set after obtaining the seed point set In the corresponding point set of mark object be split with other point sets.

Optionally, according to the seed point set, determine that multiple point concentrates the corresponding pending point set of the mark object, this is waited for Processing point set is that the corresponding point of the mark object concentrates the arbitrary point set in addition to the seed point set；

It optionally, can be with when concentrating the corresponding point set of mark object and other point sets to be split multiple point First determine the mapping relations between the corresponding pending point set of the mark object and seed point set.

It should be understood that the mapping relations between the corresponding pending point set of the mark object and seed point set can be the mark The corresponding pending point set of object and seed point set label having the same, can be determined pending by the label of seed point set Point set.

It is alternatively possible to be determined between the corresponding pending point set of the mark object and seed point set according to such as under type Mapping relations：

The spatial neighborhood relations between each point set are concentrated according to multiple point, determine the pending point set and the seed point Mapping relations between collection.

It should be understood that multiple point concentrates any one horizontal point set point set vertical with any one to may be constructed a point set It is right.

It should also be understood that it is multiple point concentrate constitute point set pair two point sets between supporting relation can use supporting degree into Row measurement, when supporting degree is more than certain threshold value, the supporting relation for indicating to constitute between two point sets of point set pair is good, or Person indicates that it is adjacent point set to constitute two point sets of point set pair.

It should also be understood that the spatial neighborhood relations between point set can be measured with the supporting degree between point set, for example, working as When multiple point concentrates the supporting degree between two point sets for constituting point set pair to be more than certain threshold value, determine between the two point sets Spatial neighborhood relations it is good, alternatively, determine the two point sets it is adjacent.

It is alternatively possible to determine that multiple point concentrates vertical point collection p according to following formula 3_iWith horizontal point set p_jThe point of composition Collect centering vertical point collection p_iWith horizontal point set p_jBetween supporting relation,

Specifically, the mapping relations between the pending point set and the seed point set can be determined according to such as under type：

It is alternatively possible to the mapping relations between the pending point set and the seed point set are determined according to formula 4,

Therefore, during the point cloud segmentation of the embodiment of the present application, which is determined by above-mentioned formula 3 and formula 4 Mapping relations between the corresponding pending point set of body and seed point set, while aspect is determined by interactive mode, and Point cloud segmentation is carried out under aspect, reduces the interactive quantity that user selects visual angle, while hoist point cloud splitting speed and accurate Property.

It is alternatively possible to as one embodiment, as shown in figure 3, the point cloud segmentation flow of this method 200 includes：

As shown in Figure 3a, it is the scoring at the visual angle of part of representative scene in a pending cloud, spherical shape represents Conventional visual angle is connected to different spherical surface positions and represents different apparent directions, and in fig. 3 a, there are 8 conventional visual angles, respectively with 1 to 8 digital representation, spherical gray scale is higher, and to represent visual angle scoring lower, for example, the scoring at No. 4 visual angles is higher than No. 1 visual angle, No. 6 The scoring at visual angle is higher than No. 3 visual angles；Taper, which represents, overlooks visual angle, and in fig. 3 a, there are 4 vertical view visual angles, use a, b, c, d respectively It indicates, the brightness of taper is higher, and to represent visual angle scoring higher, for example, the scoring at the visual angles a is higher than the visual angles d.

Optionally, when being split using Graph-Cut algorithms, commenting for conventional visual angle can be distinguished with spherical color Point, the color of taper can be used to distinguish the scoring for overlooking visual angle, for example, can be represented with being distinguished by color red to purple in chromatography Visual angle is scored from low to high, is scored it is of course also possible to be distinguished by brightness, higher, the application for example, the higher representative of brightness is scored It is not restricted to this.

As shown in Figure 3b, to input schematic diagram, it is special that the corresponding part point set of object is each marked in the input schematic diagram Different label, as each marked the White curves on object in Fig. 3 b, each point set difference for marking the corresponding White curves label of object Accordingly to mark the corresponding seed point set of object.

It is alternatively possible to a pending cloud is divided into multiple input view, for example, a pending cloud is divided into 4 input views.Optionally, an aspect is determined in each input view, and a cloud minute is carried out under aspect It cuts.

As shown in Figure 3c, it is the result schematic diagram of a point cloud segmentation, each object that marks is come out by independent Ground Split, For example, in 2,000,000 points of scene, the time for completing a Graph-Cut is about 100ms (Inteli7+GTX660).

As shown in figure 3, being a point cloud segmentation process from Fig. 3 a to 3c.

Optionally, as shown in Figures 4 to 7, it is 4 point cloud segmentation examples of this method 200, wherein the left side of every width figure For visual angle assessment result, centre is segmentation result, and right side is visual angle example.4 numbers (1,2,3 and 4) in visual angle is assessed It scores in corresponding 4 input figures higher visual angle, i.e. the aspects of 4 input views in pending cloud, in visual angle example In the corresponding view of each number correspond to the input view during left side perspective is assessed under aspect respectively.

Fig. 8 is the schematic block diagram according to the equipment 300 of the point cloud segmentation of the embodiment of the present application.As shown in figure 8, the cloud The equipment 300 of segmentation includes：

Combining unit 310, the point for treating in process points cloud merge, and obtain multiple point sets；

Sampling unit 320 obtains multiple visual angle for carrying out visual angle sampling under a multiple of viewing angles to the pending cloud In view under each visual angle, multiple visual angle is the visual angle of at least two representative scenes in the pending cloud；

Determination unit 330 is tied for assessing the view under each visual angle in multiple visual angle, and according to assessment Fruit determines aspect from multiple visual angle；

Display unit 340, the figure for after determining the aspect, showing the instruction of the view under the aspect Picture；

Acquiring unit 350, the seed point set for obtaining user's mark, the seed point set are the views under the aspect In the corresponding part point set of mark object in multiple point set；

Cutting unit 360, for according to the seed point set, the corresponding point set of mark object and its to be concentrated to multiple point Its point set is split.

Optionally, which is used for：

Optionally, which further includes：

Adjustment unit 370, for concentrating the first point set that can not merge into multiple point set when this at least two first points Number be more than third threshold value when, the second threshold is adjusted at least once, so as to which first point of multiple point set can not be merged into Collection partly or entirely merges into multiple point set.

Optionally, which is used for：

Determine that the visual angle of highest priority is the aspect.

Optionally, which is used for：

According to formulaDetermine the aspect in multiple visual angle；

Optionally, which is used for：

According to formulaDetermine the aspect in current multiple visual angles；

It should be understood that according to above and other behaviour of the modules in the equipment 300 of the point cloud segmentation of the embodiment of the present application Make and/or function respectively in order to realize the embodiment of the present application method 200 corresponding flow, for sake of simplicity, no longer superfluous herein It states.

Fig. 9 shows the schematic block diagram of the equipment 400 of point cloud segmentation provided by the embodiments of the present application, the equipment 400 packet It includes：

Memory 410, for storing program code；

Processor 420, for executing the program code in memory 410, when the program code executes, the processor 420 are used for：The point treated in process points cloud merges, and obtains multiple point sets；To the pending cloud under a multiple of viewing angles into Row visual angle samples, and obtains the view under each visual angle in multiple visual angle, multiple visual angle is at least two in the pending cloud The visual angle of a representative scene；View under each visual angle in multiple visual angle is assessed, and is tied according to assessment Fruit determines aspect from multiple visual angle；After determining the aspect, the view instruction under the aspect is shown Image；The seed point set of user's mark is obtained, which is multiple point set acceptance of the bid in the view under the aspect Note the corresponding part point set of object；According to the seed point set, to multiple point concentrate the corresponding point set of mark object with it is other Point set is split.

It should be understood that in the embodiment of the present application, which can be central processing unit (Central Processing Unit, CPU), the processor 420 can also be other general processors, digital signal processor (DSP), specially With integrated circuit (ASIC), ready-made programmable gate array (FPGA) either other programmable logic device, discrete gate or crystal Pipe logical device, discrete hardware components etc..General processor can be microprocessor or the processor can also be any normal The processor etc. of rule.

The memory 410 may include read-only memory and random access memory, and to processor 420 provide instruction and Data.The a part of of memory 410 can also include nonvolatile RAM.For example, memory 410 can also be deposited Store up the information of device type.

Optionally, which is used for：

Determine that the visual angle of highest priority is the aspect.

Optionally, which is used for：

Extract the projected area A of the point in multiple visual angle under each visual angle_pa, point comentropy A_vaWith the density A of point_pd；

According to the projected area A of the point under each visual angle_pa, point comentropy A_vaWith the density A of point_pd, determine the target Visual angle.

Optionally, which is used for：

According to formulaDetermine the aspect in multiple visual angle；

Optionally, which is used for：

According to formulaDetermine the aspect in current multiple visual angles；

It should be understood that according to processor 420 and memory 410 in the equipment 400 of the point cloud segmentation of the embodiment of the present application Above and other operation and/or function respectively in order to realize the embodiment of the present application method 200 corresponding flow, for sake of simplicity, Details are not described herein.

The embodiment of the present application provides a kind of computer readable storage medium, for storing instruction, when the instruction is calculating When being run on machine, which can be used for executing the method 200 of the point cloud segmentation of above-mentioned the embodiment of the present application.The readable medium Can be ROM or RAM, the embodiment of the present application is without limitation.

It should be understood that the terms "and/or" and " at least one of A or B ", only a kind of description affiliated partner Incidence relation, indicate may exist three kinds of relationships, for example, A and/or B, can indicate：Individualism A, exists simultaneously A and B, These three situations of individualism B.In addition, character "/" herein, it is a kind of relationship of "or" to typically represent forward-backward correlation object.

It should be understood that in each embodiment of the application, the serial number size of above-mentioned each process is not meant to execute sequence Priority, each process execution sequence should be determined by its function and internal logic, the implementation without coping with the embodiment of the present application Journey constitutes any restriction.

Those of ordinary skill in the art may realize that lists described in conjunction with the examples disclosed in the embodiments of the present disclosure Member and algorithm steps can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually It is implemented in hardware or software, depends on the specific application and design constraint of technical solution.Professional technician Each specific application can be used different methods to achieve the described function, but this realization is it is not considered that exceed Scope of the present application.

It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description, The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.

In several embodiments provided herein, it should be understood that disclosed systems, devices and methods, it can be with It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit It divides, only a kind of division of logic function, formula that in actual implementation, there may be another division manner, such as multiple units or component It can be combined or can be integrated into another system, or some features can be ignored or not executed.

The unit illustrated as separating component may or may not be physically separated, aobvious as unit The component shown may or may not be physical unit, you can be located at a place, or may be distributed over multiple In network element.Some or all of unit therein can be selected according to the actual needs to realize the mesh of this embodiment scheme 's.

In addition, each functional unit in each embodiment of the application can be integrated in a processing unit, it can also It is that each unit physically exists alone, it can also be during two or more units be integrated in one unit.

It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product It is stored in a computer read/write memory medium.Based on this understanding, the technical solution of the application is substantially in other words The part of the part that contributes to existing technology or the technical solution can be expressed in the form of software products, the meter Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be People's computer, server or network equipment etc.) execute each embodiment the method for the application all or part of step. And storage medium above-mentioned includes：USB flash disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), arbitrary access are deposited The various media that can store program code such as reservoir (Random Access Memory, RAM), magnetic disc or CD.

The above, the only specific implementation mode of the application, but the protection domain of the application is not limited thereto, it is any Those familiar with the art can easily think of the change or the replacement in the technical scope that the application discloses, and should all contain It covers within the protection domain of the application.Therefore, the protection domain of the application should be based on the protection scope of the described claims.

Claims

1. a kind of method of point cloud segmentation, which is characterized in that including：

The point treated in process points cloud merges, and obtains multiple point sets；

Visual angle sampling is carried out under a multiple of viewing angles to the pending cloud, obtains regarding under each visual angle in the multiple visual angle Figure, the multiple visual angle is the visual angle of at least two representative scenes in the pending cloud；

View under each visual angle in the multiple visual angle is assessed, and according to assessment result from the multiple visual angle Determine aspect；

After determining the aspect, the image of the view instruction under the aspect is shown；

The seed point set of user's mark is obtained, the seed point set is that multiple points described in view under the aspect are concentrated Mark the corresponding part point set of object；

According to the seed point set, divide with other point sets marking the corresponding point set of object described in the multiple point set It cuts.

2. according to the method described in claim 1, it is characterized in that, the point treated in process points cloud merges, obtain Multiple point sets, including：

The normal direction of the curvature and the consecutive points each put put according to each of described pending cloud, normal direction projection Distance and color space at least one of are estimated apart from equidistant, and the point of the pending cloud is merged into the multiple point Collection.

3. according to the method described in claim 2, it is characterized in that, the song put according to each of described pending cloud The rate and normal direction of the consecutive points each put, normal direction projector distance and color space apart from it is equidistant estimate at least The point of the pending cloud is merged into the multiple point set by one kind, including：

Choose the point that the multiple point set and curvature minimum are not yet merged into the pending cloud；

Normal direction, the normal direction projector distance of the multiple point set and the consecutive points of the point of curvature minimum are not yet merged into described in determining At least one of estimate apart from equidistant with color space, the consecutive points are the point for not yet merging into the multiple point set And/or the center of mass point of the point set of the multiple point concentration；

When the normal direction of the consecutive points, normal direction projector distance and color space apart from it is equidistant at least one of estimate be less than the When one threshold value, the point for not yet merging into the multiple point set and curvature minimum and the consecutive points are merged into described more The point set that a point is concentrated.

4. according to the method described in claim 2, it is characterized in that, the song put according to each of described pending cloud The rate and normal direction of the consecutive points each put, normal direction projector distance and color space apart from it is equidistant estimate at least The point of the pending cloud is merged into the multiple point set by one kind, including：

Normal direction, the normal direction projector distance of first point set and the consecutive points of the point of curvature minimum are not yet merged into described in determining At least one of estimate apart from equidistant with color space, the consecutive points are the point for not yet merging into first point set And/or the center of mass point of first point set；

When the normal direction of the consecutive points, normal direction projector distance and color space apart from it is equidistant at least one of estimate be less than the When one threshold value, the point for not yet merging into first point set and curvature minimum is merged at least two with the consecutive points A first point set；

Described at least two first points of concentrations are chosen not yet to merge into the multiple point set and include first points of points at most Collection；

It determines and described not yet merge into the multiple point set and include that the fitting of adjacent point set of the first most point set of points is put down At least one of normal direction, barycenter and the color histogram in face, the adjacent point set are not yet to merge into the multiple point set The point set that first point set and/or the multiple point are concentrated；

When at least one of normal direction, barycenter and color histogram of fit Plane of the adjacent point set are less than second threshold When, it not yet merges into the multiple point set by described and includes the first most point set of points and the adjacent point set and be The point set that the multiple point is concentrated；

When at least one of normal direction, barycenter and color histogram of fit Plane of the adjacent point set are more than or equal to the When two threshold values, the multiple point set is not yet merged into described in determination and includes counting the first most point sets as the multiple point The point set concentrated.

5. according to the method described in claim 4, it is characterized in that, the method further includes：

The number for the first point set that can not merge into the multiple point set is concentrated to be more than third threshold when described at least two first points When value, the second threshold is adjusted at least once, so that the first point set that can not merge into the multiple point set is part or all of Merge into the multiple point set.

6. according to any method in claim 1 to 5, which is characterized in that it is described according to the seed point set, to described The corresponding point set of object is marked described in multiple point sets to be split with other point sets, including：

According to the seed point set, determines and mark the corresponding pending point set of object described in the multiple point set, it is described to wait locating Reason point set is that the corresponding point of the mark object concentrates the arbitrary point set in addition to the seed point set；

According to the pending point set, divide with other point sets marking the corresponding point set of object described in the multiple point set It cuts.

7. according to the method described in claim 6, it is characterized in that, described according to the pending point set, to the multiple point The corresponding point set of object is marked described in collection to be split with other point sets, including：

The area of convex closure is corresponded to according to the multiple point concentration level point collection, vertical point collection corresponds to the area of convex closure, and horizontal Spatial neighborhood relations between point set and vertical point set, between determining that the multiple point concentrates two point sets for constituting point set pair Supporting relation, wherein the horizontal point set is the point set that the multiple point concentration is less than 45 ° with ground angle, the vertical point Collection is the point set that the multiple point concentration is more than or equal to 45 ° with ground angle, a horizontal point set and a vertical point set Constitute set pair；

The supporting relation between two point sets for constituting point set pair is concentrated according to the multiple point, and it is every to determine that the multiple point is concentrated Spatial neighborhood relations between a point set；

The spatial neighborhood relations between each point set are concentrated according to the multiple point, determine the pending point set and the seed Mapping relations between point set；

After determining the mapping relations between the pending point set and the seed point set, according to the pending point set, It is split with other point sets to marking the corresponding point set of object described in the multiple point set.

8. the method according to the description of claim 7 is characterized in that it is described according to it is the multiple point concentrate level point collection correspond to it is convex Area, the vertical point collection of packet correspond to the spatial neighborhood relations between the area and horizontal point set and vertical point set of convex closure, determine The multiple point concentrates the supporting relation between two point sets for constituting point set pair, including：

According to formulaDetermine that the multiple point concentrates vertical point collection p_i With horizontal point set p_jThe vertical point set p of point set centering of composition_iWith horizontal point set p_jBetween supporting relation,

Wherein, Q<p_i,p_j>Indicate the vertical point set p of composition point set pair_iWith horizontal point set p_jBetween supporting relation, p_iIt is i-th Vertical point set, p_jIt is j-th of horizontal point set, W_iIt is point set p_iThe area of corresponding convex closure, W_jIt is point set p_jThe area of corresponding convex closure, α It is predefined parameter, U (p_i,p_j) be and p_iIt is adjacent and equally support p_jVertical point set number, as U (p_i,p_j) ≠ 0 is vertical Point set p_iSupport level point set p_j, U (p_i,p_j)=0 is horizontal point set p_jSupport vertical point set p_i。

9. method according to claim 7 or 8, which is characterized in that it is described according to the multiple point concentrate each point set it Between spatial neighborhood relations, determine the mapping relations between the pending point set and the seed point set, including：

According between the seed point set and pending point set data cost and the pending point set with it is described pending Smooth cost between the adjacent point set of point set determines the mapping relations between the pending point set and the seed point set,

Wherein, the data cost is difference by the plane characteristic between pending point set and seed point set, pending point set Center and the center of nearest seed point set between space length, the color histogram between pending point set and seed point set Difference determine；The smooth cost is generated when adjacent point set belongs to a different category with pending point set by pending point set The cost of color characteristic, the fit Plane that generates when belonging to a different category of adjacent point set of pending point set and pending point set The supporting relation feature that the adjacent point set of the cost of feature, pending point set and pending point set generates when belonging to a different category Cost determines.

10. according to the method described in claim 9, it is characterized in that, it is described according to the seed point set and pending point set it Between data cost and the pending point set and the adjacent point set of the pending point set between smooth cost, determine Mapping relations between the pending point set and the seed point set, including：

Wherein, E indicates the mapping relations between the pending point set and the seed point set, E_D(p_i,v_i)=λ_DD_D(p_i,U_i)+ λ_CD_C(p_i,U_i)+λ_PD_P(p_i,U_i)+λ_SD_S(p_i,U_i), the data cost E_DIt is pending point set p_iIt is classified as label v_iCost Function, p_iIt is pending point set, v_iIt is the corresponding label of seed point collection；E_S(p_i,v_i,p_j,v_j)=λ_C′S_C(p_i,v_i,p_j,v_j)+λ_P′ S_P(p_i,v_i,p_j,v_j)+λ_S′S_S(p_i,v_i,p_j,v_j), the smooth cost E_SIt is the consecutive points of pending point set and pending point set Collect the point set pair (p of composition_i,p_j) it is classified as label (v respectively_i,v_j) cost function, p_jIt is the adjacent point set of pending point set, v_jIt is the corresponding label of consecutive points collection of seed point set；U_iIt is to be classified as label v_iAll seed point sets constitute set；D_DIt is The cost of distance feature depends on p_iCenter and the center of nearest seed point set between space length；D_CIt is the generation of color characteristic Valence depends on p_iWith the difference of the color histogram of seed point set；D_PIt is the cost of plane characteristic, depends on p_iWith seed point set Between plane characteristic (distance of normal direction and normal direction) difference；D_SIt is the cost of supporting relation feature, depends on p_iWith seed point set Between supporting relation；S_CIt is to work as p_iAnd p_jLabel (the v to belong to a different category_i≠v_j) when the cost of color characteristic that generates；S_P It is to work as p_iAnd p_jLabel (the v to belong to a different category_i≠v_j) when the cost of fit Plane feature that generates；S_SIt is to work as p_iAnd p_jBelong to Different classes of label (v_i≠v_j) when the cost of supporting relation feature that generates；λ is constant, λ_D、λ_C、λ_S、λ_C′、λ_P' and λ_S′ Constant.

11. according to any method in claims 1 to 10, which is characterized in that described to each in the multiple visual angle View under visual angle is assessed, and aspect is determined from the multiple visual angle according to assessment result, including：

Show the image of the view instruction in the multiple visual angle under each visual angle；

Obtain marking result of the user to the view under each visual angle in the multiple visual angle；

According to the marking as a result, determining the priority at the multiple visual angle；

Determine that the visual angle of highest priority is the aspect.

12. according to any method in claims 1 to 10, which is characterized in that described to each in the multiple visual angle View under visual angle is assessed, and aspect is determined from the multiple visual angle according to assessment result, including：

Extract the projected area A of the point in the view in the multiple visual angle under each visual angle_pa, point comentropy A_vaIt is close with point Spend A_pd；

According to the projected area A of the point in the view under each visual angle_pa, point comentropy A_vaWith the density A of point_pd, determine The aspect.

13. according to the method for claim 12, which is characterized in that the point in the view according under each visual angle Projected area A_pa, point comentropy A_vaWith the density A of point_pd, determine the aspect, including：

According to the marking as a result, determining the projected area A of the point in the view in the multiple visual angle under each visual angle_pa, point Comentropy A_vaWith the density A of point_pdWeight；

According to formulaDetermine the aspect in the multiple visual angle；

Wherein, V_n+1Indicate the aspect, Y_i=β^T·A_i, p_iIt is all point sets under the i of visual angle, p_nIt is all under the n of visual angle Point set, visual angle n are a upper visual angle of visual angle i, and β is the projected area A put under the i of visual angle_pa, point comentropy A_vaIt is close with point Spend A_pdWeight composition three-dimensional vector, A_iIt is the projected area A put under the i of visual angle_pa, point comentropy A_vaWith the density A of point_pd The three-dimensional vector of composition.

14. according to the method for claim 12, which is characterized in that the point in the view according under each visual angle Projected area A_pa, point comentropy A_vaWith the density A of point_pd, determine the aspect, including：

During point set is divided, the information of the point in the view in presently described multiple visual angles under each visual angle is calculated in real time Entropy A_va', and reacquire again marking result of the user to the view under each visual angle in current multiple visual angles；

According to the comentropy A of the point in the view under each visual angle calculated in real time_va' and it is described again marking as a result, determine The projected area A of the point in view in current multiple visual angles under each visual angle_pa' density the A with point_pd′；

According to formulaDetermine the aspect in current multiple visual angles；

Wherein, V_n+1Indicate the aspect, Y_i=β^T·A_i', p_iIt is all point sets under the i of visual angle, p_nIt is the institute under the n of visual angle Pointed set, visual angle n are a upper visual angle of visual angle i, and β is the projected area A put under the i of visual angle_pa', point comentropy A_va' and point Density A_pd' weight composition three-dimensional vector, A_i' it is the projected area A put under the i of visual angle_pa', point comentropy A_va' and point Density A_pdThe three-dimensional vector of ' composition.

15. a kind of equipment of point cloud segmentation, which is characterized in that including：

Combining unit, the point for treating in process points cloud merge, and obtain multiple point sets；

Sampling unit is obtained for carrying out visual angle sampling under a multiple of viewing angles to the pending cloud in the multiple visual angle View under each visual angle, the multiple visual angle is the visual angle of at least two representative scenes in the pending cloud；

Determination unit, for assessing the view under each visual angle in the multiple visual angle, and according to assessment result from Aspect is determined in the multiple visual angle；

Display unit, the image for after determining the aspect, showing the instruction of the view under the aspect；

Acquiring unit, the seed point set for obtaining user's mark, the seed point set are in the view under the aspect The corresponding part point set of object is marked in the multiple point set；

Cutting unit is used for according to the seed point set, to marking the corresponding point set of object and its described in the multiple point set Its point set is split.

16. equipment according to claim 15, which is characterized in that the combining unit is used for：

17. equipment according to claim 16, which is characterized in that the combining unit is used for：

18. equipment according to claim 16, which is characterized in that the combining unit is used for：

19. equipment according to claim 18, which is characterized in that the equipment further includes：

Adjustment unit, for concentrating the first point set that can not merge into the multiple point set when described at least two first points When number is more than third threshold value, the second threshold is adjusted at least once, so as to which first point of the multiple point set can not be merged into Collection partly or entirely merges into the multiple point set.

20. according to any equipment in claim 15 to 19, which is characterized in that the cutting unit is used for：

21. equipment according to claim 20, which is characterized in that the cutting unit is used for：

22. equipment according to claim 21, which is characterized in that the determination unit is used for：

23. the equipment according to claim 21 or 22, which is characterized in that the determination unit is used for：

24. equipment according to claim 23, which is characterized in that the determination unit is used for：

25. according to any equipment in claim 15 to 24, which is characterized in that the determination unit is used for：

Determine that the visual angle of highest priority is the aspect.

26. according to any equipment in claim 15 to 24, which is characterized in that the determination unit is used for：

27. equipment according to claim 26, which is characterized in that the determination unit is used for：

According to formulaDetermine the aspect in the multiple visual angle；

28. equipment according to claim 26, which is characterized in that the determination unit is used for：

According to formulaDetermine the aspect in current multiple visual angles；

29. a kind of computer readable storage medium, including instruction, when described instruction is run on computers, the computer Execute the method as described in any in claim 1 to 14.