CN115100272A - Prefabricated part point cloud data set manufacturing method for deep learning segmentation network - Google Patents
Prefabricated part point cloud data set manufacturing method for deep learning segmentation network Download PDFInfo
- Publication number
- CN115100272A CN115100272A CN202210721603.5A CN202210721603A CN115100272A CN 115100272 A CN115100272 A CN 115100272A CN 202210721603 A CN202210721603 A CN 202210721603A CN 115100272 A CN115100272 A CN 115100272A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- steel bar
- area
- concrete
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/64—Three-dimensional objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geometry (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Generation (AREA)
Abstract
The invention discloses a prefabricated part point cloud data set manufacturing method for a deep learning segmentation network, which comprises the following steps of: step a: generating various steel bar and concrete point clouds with marked information based on the design specification of the prefabricated part and the coordinate matrix; step b: combining the steel bars generated in the step a and the concrete point cloud to form various prefabricated part point clouds; step c: and c, inputting the point cloud of the prefabricated part in the step b into a set of automatic data preprocessing process, and outputting a data set with labeling information for deep learning segmentation network training. The method and the device have the advantages that the disadvantages that in the prior art, the time consumption and the cost are high in a method for acquiring the point cloud data set of the prefabricated part based on two-dimensional picture or three-dimensional laser scanning, and the manpower cost consumption of a manual marking method mainly utilized by the traditional point cloud data set is too large, and the accuracy degree is different from person to person are relieved.
Description
Technical Field
The invention relates to the field of prefabricated part point cloud data sets, in particular to a prefabricated part point cloud data set manufacturing method for a deep learning segmentation network.
Background
The data set is an important component of deep learning, and the development of the deep learning is promoted. In the deep learning research aspect of three-dimensional point cloud, more and more methods are proposed to solve various problems related to point cloud processing, including three-dimensional shape classification, three-dimensional target detection, three-dimensional point cloud segmentation and three-dimensional reconstruction. However, point cloud datasets related to prefabricated parts are still scarce, which limits the study of deep learning algorithms on prefabricated part point cloud processing, such as the relevant measurement or detection of prefabricated parts based on deep learning algorithms to some extent.
The collection of data sets is a difficult and time consuming part of deep learning. The most widely used practice for collecting prefabricated part point cloud data at this stage includes scanning with a laser scanner, image-based geometric methods, or converting a 3D model developed using commercial BIM software into a point cloud via various modeling software. However, these approaches are time consuming, labor intensive, and inefficient.
After the point cloud data set is obtained, the point cloud data set needs to be labeled, and because the label has semantic information, the point cloud data set is very important for developing and evaluating a deep learning algorithm. The traditional point cloud data marking mainly adopts a manual marking mode, so that the time consumption is long, the data processing labor cost is high, particularly, the manual marking needs subjective judgment, the precision is different from person to person, and the wrong marking can influence the training work of deep learning.
Disclosure of Invention
The invention aims to provide a prefabricated part point cloud data set manufacturing method facing a deep learning segmentation network aiming at the defects of the prior art.
The purpose of the invention is realized by the following technical scheme: a prefabricated part point cloud data set manufacturing method for a deep learning segmentation network comprises the following steps:
(1) determining the design sizes of concrete and steel bars in the prefabricated parts according to the design specifications of different types of prefabricated parts;
(2) calculating the area of a plane where the concrete and the steel bars can generate point clouds according to the design sizes of the concrete and the steel bars in the prefabricated part, wherein the area is in millimeter square as an area unit, introducing an eta parameter to express the integrity degree of the steel bar point clouds in the calculation process of the point number of the steel bar point clouds, and linking the area of the plane where the point clouds are generated with the point number of the point clouds which can be generated to obtain the point number of the point clouds which can be generated by the concrete and the steel bars;
(3) acquiring a coordinate matrix capable of generating the point cloud based on the geometrical characteristics of the plane of the point cloud generated by the concrete and the steel bars and the number of points capable of generating the point cloud, and generating the concrete point cloud and the steel bar point cloud according to the coordinate matrix capable of generating the point cloud and a Python environment;
(4) introducing density adjustment parameters to adjust the point densities of the concrete point cloud and the reinforcing steel bar point cloud;
(5) after the point densities of the concrete point cloud and the reinforcing steel bar point cloud are adjusted, adding respective marking information to the concrete point cloud and the reinforcing steel bar point cloud to generate various reinforcing steel bars and concrete point clouds with marking information;
(6) combining the various steel bars with the marked information generated in the step (5) with the concrete point cloud to form various prefabricated part point clouds;
(7) and (4) sequentially carrying out point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling on the prefabricated part point cloud formed in the step (6) to obtain a prefabricated part point cloud data set with marked information.
Further, the step (1) specifically includes the following sub-steps:
(1.1) designing concrete a with different design sizes according to the design specifications of different types of prefabricated parts, wherein the concrete a is a cuboid, and the length, the width and the height of the concrete a are respectively l a 、w a 、h a The plane of the concrete a capable of generating the point cloud comprises five parts, namely a top plane, a left plane, a right plane, a front plane and a rear plane of a cuboid;
(1.2) designing steel bars with different design sizes according to the design specifications of different types of prefabricated parts, wherein the steel bars comprise straight steel bars, hook steel bars and U-shaped steel bars;
the design size of the straight steel bar is as follows: the straight reinforcing steel bar i consists of a cone and a cylinder, and the radiuses of the upper bottom surface and the lower bottom surface of the cone are r respectively i2 、r i1 Height of h i1 (ii) a Height of cylinder is h i2 Radius of the bottom surface is r i2 (ii) a The plane on which the straight steel bar i can generate the point cloud comprises three parts, namely a circular truncated cone side surface, a cylinder side surface and a cylinderAn upper body bottom surface;
the design size of the hook steel bar is as follows: the hook steel bar j consists of a circular truncated cone, a first cylinder, a partial circular ring body and a second cylinder, and the radiuses of the upper bottom surface and the lower bottom surface of the circular truncated cone are r respectively j2 、r j1 Height of h j1 (ii) a The height of the first cylinder is h j2 Radius of the bottom surface is r j2 (ii) a The radius of the circular section, i.e. the generatrix, of the partial torus is r j2 The distance from the center of the generatrix to the axis of the ring is R j The central angle is alpha, and the alpha represents the bending angle of the steel bar of the hook steel bar; the radius of the bottom surface of the second cylinder is r j2 Height is l j (ii) a The plane on which the hook steel bar j can generate the point cloud comprises five parts, namely a circular truncated cone side surface, a first cylinder side surface, a partial circular ring side surface, a second cylinder side surface and a second cylinder upper bottom surface;
the design size of the U-shaped steel bar is as follows: the U-shaped steel bar k consists of a first circular truncated cone body, a first cylinder, a half circular ring body, a second cylinder and a second circular truncated cone body, wherein the radiuses of the upper bottom surface and the lower bottom surface of the first circular truncated cone body are r respectively k2 、r k1 Height of h k1 (ii) a The height of the first cylinder is h k1 Radius of the bottom surface is r k2 (ii) a The radius of the circular section, i.e. the generatrix, of the half circular ring body is r k2 The distance from the center of the generatrix to the axis of the ring is R k The central angle is pi; height of the second cylinder is h k1 Radius of the bottom surface is r k2 (ii) a The upper and lower bottom radii of the second circular truncated cone are r k2 、r k1 Height of h k1 (ii) a The plane on which the U-shaped steel bar kj can generate the point cloud comprises five parts, namely a first circular truncated cone side face, a first cylinder side face, a semi-circular-ring side face, a second cylinder side face and a second circular truncated cone side face.
Further, the step (2) specifically comprises the following sub-steps:
(2.1) calculating the area of a plane capable of generating the point cloud by the concrete, wherein the area unit is millimeter square, and the area is related to the number of points, and the number of the points capable of generating the point cloud by the concrete is equal to the area of the plane capable of generating the point cloud by the concrete; the calculation formula of the number of points of the concrete capable of generating the point cloud is as follows:
wherein a is the concrete type number; s 1 、S 2 、S 3 、S 4 、S 5 The area of the top plane, the area of the left side plane, the area of the right side plane, the area of the front side plane and the area of the rear side plane of the concrete a are respectively taken as the area unit of millimeter square; s a The area of the plane for which the point cloud can be generated for the concrete a, and the number of points of the point cloud correspondingly generated for the concrete a; s concrete For the number of points of all the concrete point clouds in the prefabricated part point cloud, the concrete with different design sizes in b is shared, n a The amount of concrete a;
(2.2) calculating the area of a plane of the steel bar capable of generating the point cloud, wherein the area unit is millimeter square, introducing an eta parameter to express the integrity degree of the steel bar point cloud, and then calculating the number of points of the steel bar capable of generating the point cloud; eta is obtained by dividing theta by 2 pi, and theta is a central angle corresponding to an arc formed by projecting an area with point cloud distribution on the side part of the cylinder in the steel bar to the bottom plane of the cylinder;
number of points S of steel bar point cloud in prefabricated part rebar Calculating the point number S of the point clouds of straight steel bars, hook steel bars and U-shaped steel bars straight 、S curved 、S U-shaped Adding to obtain;
a) number of points S of straight steel bar capable of generating point cloud straight The calculation process of (2):
calculating the area of a plane of the point cloud generated by the straight steel bars, wherein the unit of the area is millimeter square, and introducing eta i The parameters relate the area to the number of points, the number of points of the point cloud generated by the straight steel bars is equal to the product of the area of the plane of the point cloud generated by the straight steel bars and eta i (ii) a The calculation formula of the number of points of the point cloud generated by the straight steel bars is as follows:
wherein i is the type number of the straight steel bar; s iA 、S iB 、S iC The area of the side surface of the circular truncated cone body, the area of the side surface of the cylinder and the area of the upper bottom surface of the cylinder which are respectively the straight reinforcing steel bar i all take millimeter square as an area unit; s i Number of points, eta, of point clouds correspondingly generated for straight bars i i Representing the integrity degree of a reinforcing steel bar point cloud generated by corresponding to the straight reinforcing steel bar i; s straight For the number of points of all the point clouds of the straight reinforcing steel bars in the prefabricated part point clouds, there are o kinds of straight reinforcing steel bars with different design sizes, and n i The number of the straight steel bars i;
b) number of points S of hook steel bar capable of generating point cloud curved The calculation process of (2):
calculating the area of the plane of the hook steel bar capable of generating point cloud, wherein the unit of the area is millimeter square, and introducing eta j The parameters relate the area to the number of points, the number of points of the point cloud generated by the hook steel bar is equal to the product of the area of the plane of the point cloud generated by the hook steel bar and eta j (ii) a The calculation formula of the point number of the point cloud generated by the hook steel bar is as follows:
wherein j is the type number of the hook steel bar; s jD 、S jE 、S jF 、S jG 、S jH The area of the side surface of the circular truncated cone body of the hook steel bar j, the area of the side surface of the first cylinder body, the area of the side surface of the partial circular ring body, the area of the side surface of the second cylinder body and the area of the upper bottom surface of the second cylinder body are respectively in unit of millimeter square; s j Number of points, η, of point clouds correspondingly generated for the hook reinforcement j j Representing the integrity degree of a steel bar point cloud generated by the hook steel bar j correspondingly; s. the curved For the point number of all the hook steel bar point clouds in the prefabricated part point cloud, p hook steel bars with different design sizes are shared, and n j The number of the hooked steel bars j;
c) number of points S of point cloud can be generated by U-shaped steel bar U-shaped The calculation process of (2):
calculating the point cloud that can be generated by the U-shaped reinforcementThe area of the plane is in unit of millimeter square, and a parameter eta is introduced k The area is related to the number of points, and the number of points of the point cloud generated by the U-shaped steel bar is equal to the product of the area of a plane of the point cloud generated by the U-shaped steel bar and eta k (ii) a The calculation formula of the number of points of the point cloud generated by the U-shaped steel bar is as follows:
wherein k is the type number of the U-shaped steel bar; s kM 、S kN 、S kO 、S kP And S kQ The areas of the side surface of the first circular truncated cone body, the side surface of the first cylinder, the side surface of the half circular ring body, the side surface of the second cylinder and the side surface of the second circular truncated cone body of the U-shaped steel bar k are respectively in unit of millimeter square; s k Number of points, η, of point clouds correspondingly generated for the U-shaped reinforcement k k Representing the integrity degree of a reinforcing bar point cloud generated by the U-shaped reinforcing bar k; s U-shaped For the number of points of all the U-shaped steel bar point clouds in the prefabricated part point cloud, q U-shaped steel bars with different design sizes are arranged, and n k The number of the U-shaped reinforcing bars k.
Further, the step (6) specifically comprises the following sub-steps:
(6.1) selecting a certain amount and variety of steel bar point clouds and concrete point clouds required by the generated prefabricated part;
(6.2) performing space operations such as rotation, translation and the like on the steel bar point cloud selected in the step (6.1) to enable the steel bar point cloud and the concrete point cloud to be combined into a pattern of a required prefabricated part;
and (6.3) removing the point cloud on the surface of the concrete covered by the bottom of the steel bar to generate the point cloud of the prefabricated part.
Further, the point cloud normalization in the step (7) specifically includes: and calculating the average value of the x, y and z coordinates of all points in the prefabricated part point cloud to be used as a centroid coordinate, then calculating the distance d from the point farthest from the centroid to the centroid, subtracting the centroid coordinate from the coordinates of all the points, and dividing the centroid coordinate by d to obtain the normalized prefabricated part point cloud coordinate.
Further, the enhancing of the point cloud data in the step (7) specifically includes: and (4) performing three transformations of rotation, dithering and translation on the normalized prefabricated part point cloud coordinate to obtain the prefabricated part point cloud coordinate after data enhancement.
Further, the point cloud cutting in the step (7) specifically comprises: and calculating the coordinate ranges of the prefabricated part point cloud coordinates after data enhancement in the x dimension, the y dimension and the z dimension, dividing the coordinates in the three dimensions according to a certain threshold range, and realizing the division of the prefabricated part point cloud after data enhancement based on the division result.
Further, the point cloud downsampling in the step (7) specifically comprises: and performing down-sampling on the point cloud of the segmented prefabricated part by adopting a curvature-based sampling method.
The invention has the beneficial effects that:
(1) the point cloud point density of subcomponents (steel bars and concrete) in the prefabricated part can be respectively customized;
(2) the automatic labeling of the steel bar point cloud and the concrete point cloud is realized, the problem that the precision in the traditional data set labeling is different from person to person is solved, and the labeling precision is ensured;
(3) the data preprocessing reduces the number of points of a single sample in a data set, retains enough geometric characteristic points and the relative integrity of the reinforcing steel bar data, and can accelerate the data processing speed of deep learning;
(4) prefabricated part point cloud data sets of various types and shapes can be generated in a short time.
Drawings
FIG. 1 is a flow chart of a prefabricated part point cloud data set manufacturing method for a deep learning segmentation network provided by the invention;
FIG. 2 is a schematic diagram of the design dimensions of concrete;
fig. 3 is a schematic diagram of the design dimensions of all the steel bars, wherein fig. 3(a) is a schematic diagram of the design dimensions of straight steel bars, fig. 3(b) is a schematic diagram of the design dimensions of hook steel, and fig. 3(c) is a schematic diagram of the design dimensions of U-shaped steel bars;
fig. 4 is a schematic view of the integrity of all the rebars, wherein fig. 4(a) is a schematic view of a straight rebar with 100% integrity, fig. 4(b) is a schematic view of a straight rebar with 50% integrity, fig. 4(c) is a schematic view of a hooked rebar with 100% integrity, fig. 4(d) is a schematic view of a hooked rebar with 50% integrity, fig. 4(e) is a schematic view of a U-shaped rebar with 100% integrity, and fig. 4(f) is a schematic view of a U-shaped rebar with 50% integrity;
fig. 5 is a schematic diagram of a prefabricated part point cloud formed by combining a reinforcing steel bar point cloud and a concrete point cloud generated in the invention, wherein fig. 5(a) is a schematic diagram of a prefabricated column point cloud, fig. 5(b) is a schematic diagram of a prefabricated beam point cloud, fig. 5(c) is a schematic diagram of a prefabricated slab point cloud, and fig. 5(d) is a schematic diagram of a prefabricated wall point cloud;
fig. 6 is a prefabricated component point cloud schematic diagram after point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling processing, wherein fig. 6(a) is a prefabricated column point cloud schematic diagram after point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling processing, fig. 6(b) is a prefabricated beam point cloud schematic diagram after point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling processing, fig. 6(c) is a prefabricated plate point cloud schematic diagram after point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling processing, and fig. 6(d) is a prefabricated wall point cloud schematic diagram after point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling processing.
Detailed Description
For purposes of promoting an understanding of the objects, aspects and advantages of the invention, reference will now be made to the following detailed description of the embodiments taken in conjunction with the accompanying drawings, it being understood that the specific embodiments described herein are illustrative of the invention and are not intended to be exhaustive. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, are within the scope of the present invention.
Example 1
As shown in fig. 1, the invention provides a method for manufacturing a prefabricated part point cloud data set for a deep learning segmentation network, which comprises the following steps:
(1) determining the design sizes of concrete and steel bars in the prefabricated parts according to the design specifications of different types of prefabricated parts;
the design sizes of a concrete part and a steel bar part in the prefabricated part are correspondingly determined according to the design specifications of different types of prefabricated parts, and then the marked concrete point cloud and the marked steel bar point cloud are generated. The design specifications include the shape of the concrete portion of the prefabricated member and the type, number and distribution position of the reinforcing bars. The different types of prefabricated components comprise prefabricated columns, prefabricated beams, prefabricated plates and prefabricated walls.
The step (1) comprises the following substeps:
(1.1) designing concrete a with different design sizes according to the design specifications of different types of prefabricated parts;
the concrete a is a cuboid, the concrete parts of the prefabricated columns and the prefabricated beams are strip-shaped cuboids, and the concrete parts of the prefabricated plates and the prefabricated walls are sheet-shaped cuboids. The length, width and height of the concrete a are respectively a 、w a 、h a Is represented by a Is the length of concrete a, w a Is the width of the concrete a, h a Designing length, width and height (l) with different numerical value proportions for the height of the concrete a according to different types of prefabricated parts a 、w a 、h a ) (ii) a Since the concrete bottom is not visible in the actual three-dimensional scan, the bottom plane of the concrete is deleted and the plane in which the concrete a can generate the point cloud includes five parts, as shown in fig. 2, which are the top planes of the cuboid (l) a *w a ) Left side plane (l) a *h a ) Right side plane (l) a *h a ) Front side plane (w) a *h a ) And a rear plane (w) a *h a )。
(1.2) designing steel bars with different design sizes according to the design specifications of different types of prefabricated parts, wherein the steel bars comprise straight steel bars, hook steel bars and U-shaped steel bars;
among the different types of prefabricated component, prefabricated post generally only contains straight reinforcing bar, and the precast beam contains straight reinforcing bar and crotch reinforcing bar, and the prefabricated plate contains straight reinforcing bar and crotch reinforcing bar, and the prefabricated wall contains the U-shaped reinforcing bar, so the design has three kinds of reinforcing bars altogether: straight reinforcing bars, hooked reinforcing bars and U-shaped reinforcing bars.
a) The design size of the straight steel bar is as follows:
the straight steel bar i consists of a cone and a cylinder, and the design size of the straight steel bar is shown in fig. 3 (a); the radius of the upper and lower bottom surfaces of the truncated cone body is r i2 、r i1 Height of h i1 (ii) a Height of cylinder is h i2 The lower bottom surface of the cylinder coincides with the upper bottom surface of the cone, namely the radius of the bottom surface of the cylinder is r i2 (ii) a The plane of the straight steel bar i capable of generating the point cloud comprises three parts which are respectively the side surfaces (S) of the circular truncated cone A ) Cylinder side (S) B ) And the upper bottom surface (S) of the cylinder C ) In fig. 3(a), A, B, C parts respectively correspond to the side surface of the circular truncated cone, the side surface of the cylinder and the upper bottom surface of the cylinder; r is i1 、r i2 、h i1 、h i2 The value of (b) is self-defined based on the design specification, wherein i is the type serial number of the steel bar (different serial numbers indicate different design sizes of the straight steel bar).
b) The design size of the hook steel bar is as follows:
the hook steel bar j consists of a circular truncated cone, a first cylinder, a partial circular ring body and a second cylinder, and the design size of the hook steel bar is shown in figure 3 (b); the radius of the upper and lower bottom surfaces of the truncated cone is r j2 、r j1 Height of h j1 (ii) a The height of the first cylinder is h j2 The lower bottom surface of the first cylinder coincides with the upper bottom surface of the truncated cone, i.e. the radius of the bottom surface of the first cylinder is r j2 (ii) a The circular section of the partial circular ring body, i.e. the radius of the generating circle is r j2 The distance from the center of the generatrix to the axis of the ring is R j The central angle is alpha (alpha represents the bending angle of the steel bar of the hook steel bar); the radius of the bottom surface of the second cylinder is r j2 High is l j . The plane on which the hook steel bar j can generate the point cloud comprises five parts which are the side surfaces (S) of the circular truncated cone respectively D ) First cylinder side (S) E ) Partial ring body side (S) F ) A second cylinder side (S) G ) And the upper bottom surface (S) of the second cylinder H ) In FIG. 3(b), D, E, F, G and H correspond to the circular truncated cone side surface, the first cylindrical side surface, and the partial circular truncated cone side surface, respectivelyThe surface, the side surface of the second cylinder and the upper bottom surface of the second cylinder; r is j1 、r j2 、h j1 、h j2 、α、R j And l j The numerical value of (b) is self-defined based on the design specification, wherein j is the type serial number of the steel bar (different serial numbers indicate different design sizes of the hook steel bar).
c) The design size of the U-shaped reinforcing steel bar is as follows:
the U-shaped steel bar k consists of a first truncated cone, a first cylinder, a half circular ring, a second cylinder and a second truncated cone, and the design size of the U-shaped steel bar is shown in figure 3 (c); the radius of the upper and lower bottom surfaces of the first circular truncated cone body is r k2 、r k1 Height of h k1 (ii) a The height of the first cylinder is h k1 The lower bottom surface of the first cylinder coincides with the upper bottom surface of the truncated cone, i.e. the radius of the bottom surface of the first cylinder is r k2 (ii) a The radius of the circular section, i.e. the generatrix, of the half circular ring body is r k2 The distance from the center of the generatrix to the axis of the ring is R k The central angle is pi (the central angle is pi indicates that the circular ring is a half circular ring body); the height of the second cylinder is h k1 The radius of the bottom surface of the first cylinder is r k2 (ii) a The upper and lower bottom radii of the second circular truncated cone are r k2 、r k1 Height of h k1 . The plane of the U-shaped steel bar kj capable of generating the point cloud comprises five parts which are respectively the side surface (S) of the first round platform body M ) First cylinder side (S) N ) Side surface of half circular ring body (S) O ) A second cylinder side (S) P ) And a second circular truncated cone side surface (S) Q ) In fig. 3(c), M, N, O, P and Q parts correspond to a first circular truncated cone side surface, a first cylinder side surface, a half circular ring side surface, a second cylinder side surface and a second circular truncated cone side surface, respectively; r is k1 、r k2 、h k1 、h k2 And R k The value of (a) is self-defined based on the design specification, wherein k is the type serial number of the steel bar (different serial numbers indicate different design sizes of the U-shaped steel bar).
(2) Calculating the area of a plane where the concrete and the steel bars can generate point cloud according to the design size of the concrete and the steel bars in the prefabricated part, wherein the area is in the unit of millimeter square, introducing eta parameters to express the integrity degree of the steel bar point cloud in the calculation process of the point number of the steel bar point cloud, and associating the area of the plane where the point cloud is generated with the point number of the point cloud which can be generated to obtain the point number of the point cloud which can be generated by the concrete and the steel bars;
(2.1) calculating the area of a plane capable of generating the point cloud by the concrete, wherein the area unit is millimeter square, and the area is related to the number of points, and the number of the points capable of generating the point cloud by the concrete is equal to the area of the plane capable of generating the point cloud by the concrete; the calculation formula of the number of points of the concrete capable of generating the point cloud is as follows:
wherein a is the concrete type number (different numbers indicate different design sizes of concrete), and S is 1 、S 2 、S 3 、S 4 、S 5 Top plane area, left plane area, right plane area, front plane area and rear plane area (S) of concrete a 1 、S 2 、S 3 、S 4 、S 5 Both in units of area of millimeters squared); s a The area of a plane capable of generating point clouds for the concrete a and the number of points of the point clouds correspondingly generated for the concrete a; s concrete For the number of points of all the concrete point clouds in the prefabricated part point cloud, the concrete with different design sizes in b is shared, n a The amount of concrete a.
(2.2) calculating the area of a plane where the reinforcing steel bars can generate point clouds, wherein the area unit is millimeter square, and because the reinforcing steel bar point clouds are difficult to be completely scanned in the process of obtaining the point clouds through actual three-dimensional scanning, an eta parameter is introduced in the design to represent the integrity degree of the reinforcing steel bar point clouds, and then the number of points where the reinforcing steel bars can generate the point clouds is calculated; eta is obtained by dividing theta by 2 pi, and theta is a central angle corresponding to an arc formed by projecting an area with point cloud distribution on the side part of the cylinder in the steel bar to the bottom plane of the cylinder.
Number of points S of steel bar point cloud in prefabricated part rebar Calculating the point number S of the point clouds of straight steel bars, hook steel bars and U-shaped steel bars straight 、S curved 、S U-shaped And adding the two components to obtain the product.
a) Number of points S of straight steel bar capable of generating point cloud straight The calculation process of (2):
calculating the area of a plane of the point cloud generated by the straight steel bars, wherein the unit of the area is millimeter square, and introducing eta i The parameter relates the area to the number of points, the number of points of the point cloud generated by the straight steel bar is equal to the area of the plane of the point cloud generated by the straight steel bar multiplied by eta i (ii) a The calculation formula of the number of points of the point cloud generated by the straight steel bars is as follows:
wherein i is the type number of the straight steel bar (different numbers indicate different design sizes of the straight steel bars), and S iA 、S iB 、S iC The area of the side surface of the truncated cone, the area of the side surface of the cylinder and the area of the upper bottom surface of the cylinder of the straight steel bar i (S) iA 、S iB 、S iC Both in units of area of millimeters squared); s i Number of points, η, of point clouds generated for straight bars i i Representing the integrity degree of a reinforcing steel bar point cloud generated by corresponding to the straight reinforcing steel bar i; s. the straight For the number of points of all the point clouds of the straight reinforcing steel bars in the prefabricated part point clouds, there are o kinds of straight reinforcing steel bars with different design sizes, and n i The number of straight bars i.
b) Number of points S of hook steel bar capable of generating point cloud curved The calculation process of (2):
calculating the area of the plane of the hook steel bar capable of generating point cloud, wherein the unit of the area is millimeter square, and introducing eta j The parameters relate the area to the number of points, the number of points of the point cloud generated by the hook steel bar is equal to the product of the area of the plane of the point cloud generated by the hook steel bar and eta j (ii) a The calculation formula of the point number of the point cloud generated by the hook steel bar is as follows:
wherein j is the type number of the hook steel bar (different serial numbers indicate different design sizes of the hook steel bars), and S jD 、S jE 、S jF 、S jG 、S jH Respectively the side surface area of the circular truncated cone body, the side surface area of the first cylinder body, the side surface area of the partial torus, the side surface area of the second cylinder body and the upper bottom surface area (S) of the second cylinder body jD 、S jE 、S jF 、S jG 、S jH Both in units of area of millimeters squared); s. the j Number of points, η, of point clouds correspondingly generated for the hook reinforcement j j Representing the integrity degree of a steel bar point cloud generated by the hook steel bar j correspondingly; s curved For the number of all the points in the point cloud of the hook reinforcing steel bars in the point cloud of the prefabricated part, p types of hook reinforcing steel bars with different design sizes are provided, and n j Is the number of hooked rebar j.
c) Number of points S of point cloud can be generated by U-shaped steel bar U-shaped The calculation process of (2):
calculating the area of a plane of the point cloud generated by the U-shaped steel bar, wherein the unit of the area is millimeter square, and introducing a parameter eta k The area is related to the number of points, and the number of points of the point cloud generated by the U-shaped steel bar is equal to the product of the area of a plane of the point cloud generated by the U-shaped steel bar and eta k (ii) a The calculation formula of the number of points of the point cloud generated by the U-shaped steel bar is as follows:
wherein k is the type number of the U-shaped steel bar (different serial numbers indicate different design sizes of the U-shaped steel bar), and S kM 、S kN 、S kO 、S kP And S kQ The side surface area of the first circular truncated cone body, the side surface area of the first cylinder body, the side surface area of the half circular ring body, the side surface area of the second cylinder body and the side surface area S of the second circular truncated cone body of the U-shaped steel bar k are respectively kM 、S kN 、S kO 、S kP And S kQ Both in units of area of millimeters squared); s k Number of points, η, of point clouds correspondingly generated for the U-shaped reinforcement k k Steel bar point for representing U-shape steel bar k corresponding generationDegree of integrity of the cloud; s U-shaped For the number of points of all the U-shaped steel bar point clouds in the prefabricated part point cloud, q U-shaped steel bars with different design sizes are arranged, and n k The number of the U-shaped reinforcing bars k.
Obtaining the point number of the steel bar point cloud in the prefabricated part as S rebar ，S rebar ＝S straight +S curved +S U-shaped ，S rebar The total number of all points of the point cloud of the straight steel bars, the hook steel bars and the U-shaped steel bars in the point cloud of the prefabricated part is equal to the area value (the area calculated by taking millimeter square as a unit) corresponding to a distribution area of the point cloud of the steel bars.
(3) Acquiring a coordinate matrix capable of generating the point cloud based on the geometrical characteristics of the plane of the point cloud generated by the concrete and the steel bars and the number of points capable of generating the point cloud, and generating the concrete point cloud and the steel bar point cloud according to the coordinate matrix capable of generating the point cloud and the Python environment;
in this embodiment, taking the cylindrical side surface of the straight steel bar i as an example, the number of points on the cylindrical side surface of the straight steel bar i that can generate the point cloud is η i S iB The geometric characteristic of the side surface of the cylinder of the straight steel bar i is an arc curved surface, a corresponding coordinate matrix capable of generating point clouds is generated for the side surface of the cylinder of the straight steel bar i, and a matrix generation formula is as follows:
wherein n is the number of points of the point cloud generated by the side surface of the cylinder of the straight steel bar i, x s ，y s ，z s Is the three-dimensional coordinate of the s-th point in the reinforcing bar point cloud s X, Y, Z are matrixes formed by three-dimensional coordinates of point clouds on the side face of the cylinder of the straight steel bar i respectively, wherein the random numbers are generated by a random function and are between 0 and 1, and A is a coordinate matrix of the point clouds generated on the side face of the cylinder of the straight steel bar i;
and then generating a coordinate matrix of the point cloud according to the cylindrical side surface of the straight steel bar i and generating a cylindrical side surface point cloud of the steel bar i in a Python environment.
Other parts (plane or curved surface) are the same as the steps of generating point clouds on the side surfaces of the cylinders of the steel bars i, and finally various concrete point clouds and steel bar point clouds can be generated.
Fig. 4 is a schematic view of the integrity of all the rebars, wherein fig. 4(a) is a schematic view of a straight rebar with 100% of integrity, fig. 4(b) is a schematic view of a straight rebar with 50% of integrity, and the A, B, C part in fig. 4(a) and fig. 4(b) corresponds to the A, B, C part in fig. 3 (a).
Fig. 4(c) is a schematic view of a 100% complete hook bar, fig. 4(d) is a schematic view of a 50% complete hook bar, and portions D, E, F, G and H in fig. 4(c) and 4(d) correspond to portions D, E, F, G and H in fig. 3 (b).
Fig. 4(e) is a schematic view of a U-shaped rib having a degree of integrity of 100%, fig. 4(f) is a schematic view of a U-shaped rib having a degree of integrity of 50%, and M, N, O, P and Q portions in fig. 4(e) and 4(f) correspond to M, N, O, P and Q portions in fig. 3 (c).
In engineering practice, the horizontal or vertical arrangement of the steel bars can cause the defect positions of the steel bar point clouds obtained by scanning to be different. Therefore, the method sets different defect patterns by adjusting the distribution range (coordinate matrix) of the points for the same integrity degree.
(4) Introducing density adjustment parameters to adjust the point densities of the concrete point cloud and the reinforcing steel bar point cloud;
for a single prefabricated part point cloud, the number of points of the concrete point cloud and the steel bar point cloud is greatly different, and the method belongs to the category of unbalanced learning; inspired by unbalanced data research and sampling strategy in deep learning, the adjustment parameter w for adjusting the density of the concrete point cloud in the prefabricated part is introduced c And a steel bar point cloud density adjusting parameter w r With the unbalanced distribution in the compensation preparation point cloud, can be better used to the deep learning training, specifically calculate to be:
wherein S is concrete And S rebar For adjusting the number of points of the concrete point cloud and the number of points of the reinforcing steel bar point cloud before adjustment，n c And n r For the number of points of the adjusted concrete point cloud and the number of points of the steel bar point cloud, S c And S r Is the area, rho, corresponding to the distribution area of the concrete point cloud and the reinforcing bar point cloud c And ρ r The point density of the adjusted concrete point cloud and the point density of the steel bar point cloud are obtained, and mu is the ratio of the point density of the adjusted steel bar point cloud and the point density of the concrete point cloud; by adjusting w r And w c The values of (a) can be obtained as different mu, and different mu have different influences on training of a deep learning component segmentation network (verified in a PointNet + + network, the accuracy of the obtained verification sets is 0.98181, 0.98870, 0.98327, 0.98195 and 0.97037 respectively for training of prefabricated part point cloud data sets with density ratios of 1, 2, 3, 4 and 5 respectively, the accuracy reaches the peak value when the density ratio is 2 and is highest, and the accuracy gradually decreases after the density ratio is more than 2).
(5) After the point densities of the concrete point cloud and the reinforcing steel bar point cloud are adjusted, adding respective marking information to the concrete point cloud and the reinforcing steel bar point cloud to generate various reinforcing steel bars and concrete point clouds with marking information; the labeling information is category information.
The method for adding respective labeling information to the concrete point cloud and the reinforcing steel bar point cloud is as follows:
in the invention, the mark is added by an object instead of a point cloud marking tool (such as point-closed-annotation-tool or magnetic-sectional-editor); adding labels on points of the reinforcing steel bar point cloud or the concrete point cloud before the reinforcing steel bar point cloud and the concrete point cloud are combined into the prefabricated part point cloud can be regarded as a marking step; for example, "u" and "v" are added to the concrete point cloud coordinate and the steel bar point cloud coordinate respectively, and then the matrix dimension is increased to serve as a label feature, which is equivalent to that the data marking is completed before the prefabricated point cloud is manufactured; therefore, compared with the traditional labeling process, the method avoids human errors and greatly shortens the time.
(6) Combining the various steel bars with the marked information generated in the step (5) with the concrete point cloud to form various prefabricated part point clouds;
(6.1) selecting a certain amount and variety of steel bar point clouds and concrete point clouds required by the generated prefabricated part;
the prefabricated components comprise prefabricated columns, prefabricated beams, prefabricated plates and prefabricated walls.
For example, if a prefabricated column point cloud with four exposed straight steel bars needs to be manufactured, selecting a concrete point cloud and four straight steel bar point clouds as shown in fig. 5 (a); if a precast beam point cloud with five straight steel bars exposed at two ends and twenty hook steel bars exposed at the top needs to be manufactured, selecting a concrete point cloud, ten straight steel bar point clouds and twenty hook steel bar point clouds as shown in fig. 5 (b); the prefabricated panels and the prefabricated walls are similar as shown in fig. 5(c) and 5 (d).
(6.2) performing space operations such as rotation, translation and the like on the steel bar point cloud selected in the step (6.1) to enable the steel bar point cloud and the concrete point cloud to be combined into a pattern of a required prefabricated part;
for the selected concrete point cloud and the selected steel bar point cloud, the central position of the initial bottom plane is at the origin of coordinates, the positions of the concrete point cloud are kept different, different rotation angles and translation distances are executed on each steel bar point cloud, the steel bar point cloud is positioned at the designed position, and the root of the steel bar point cloud is attached to the surface of the concrete point cloud.
(6.3) removing the point cloud on the concrete surface covered by the bottom of the steel bar to generate a prefabricated part point cloud:
after the step (6.2) is finished, the root of the point cloud of the reinforcing steel bars covers some redundant points belonging to the surface of the concrete; based on the characteristics that the interior of the real scanning point cloud is hollow and only the geometric features of the surface of the object are obtained, executing a KD-tree algorithm to remove the redundant points, so that the point cloud of the combined prefabricated part is closer to the characteristics of the real scanning point cloud; fig. 5(a) - (d) are respectively prefabricated column, beam, plate and wall point clouds after the initial fabrication.
(7) And (4) carrying out point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling on the prefabricated part point cloud formed in the step (6) to obtain a prefabricated part point cloud data set with marked information.
The prefabricated part point cloud data set with the labeled information is used for deep learning segmentation network training.
Point cloud normalization, namely selecting an effective standard coordinate system taking a data centroid as a center; the point cloud normalization specifically comprises: and calculating the average value of the x, y and z coordinates of all points in the prefabricated part point cloud to be used as a centroid coordinate, then calculating the distance d from the point farthest from the centroid to the centroid, subtracting the centroid coordinate from the coordinates of all the points, and dividing the centroid coordinate by d to obtain the normalized prefabricated part point cloud coordinate. The point cloud normalization can accelerate the data processing speed, the gradient descent solving speed and the convergence speed of a deep learning model of a later deep learning algorithm;
the point cloud normalization calculation formula is as follows:
wherein x is c 、y c 、z c Respectively the calculated x, y and z coordinates of the mass center, t is the point number of the point cloud midpoint of the prefabricated part, and x g 、y g 、z g X, y and z coordinates of the g point in the prefabricated part point cloud, x h 、y h 、z h Respectively subtracting the coordinates of the center of mass from the x, y and z coordinates of the g point in the prefabricated member point cloud, wherein d is the distance from the point farthest from the center of mass to the center of mass, and x is the distance from the point farthest from the center of mass to the center of mass new 、y new 、z new Respectively representing x, y and z coordinates of the g point in the prefabricated part point cloud after normalization.
The point cloud data enhancement is to process the three-dimensional coordinates of points, and the robustness of a deep learning network generally depends on the availability and diversity of a data set. In order to enable the data set to be more comprehensive and reduce the difference between the training data set and the testing data set, data enhancement is needed, and by comprehensively amplifying the diversity of the synthetic data set, overfitting of a later-stage deep learning model is avoided and the generalization capability of a deep learning network is improved. Considering the properties of the three-dimensional point cloud, the point cloud data enhancement is specifically as follows: the normalized prefabricated component point cloud coordinate is subjected to three transformations of rotation, shaking and translation to obtain the prefabricated component point cloud coordinate after data enhancement so as to enhance deep learning network training,
the calculation process of point cloud data enhancement is as follows:
P′＝P*U+V+W
wherein P is a normalized prefabricated component point cloud three-dimensional coordinate matrix with dimension t x 3(t is the number of points of the normalized prefabricated component point cloud); u is a rotation matrix of 3 x 3, the size of three rows of numerical values is generated by a random function, and a rotation effect is generated on point cloud; v is a jitter matrix of t x 3, and the row values are not equal, so that jitter effects with different amplitudes are generated for each point; w is a translation matrix of t x 3, the numerical values of all rows are equal, and the integral translation effect is generated on point clouds; and P' is a prefabricated part point cloud three-dimensional coordinate matrix after data enhancement.
Point cloud cutting: for deep learning of three-dimensional point cloud, too many points in a single sample can influence the data processing speed and the convergence speed of a model, while too few points can cause key point loss, and the sample characteristics are difficult to capture. In order to reduce the number of points in a single sample, the data processing speed and the convergence speed of a model are prevented from being influenced by excessive points, and meanwhile, the relative integrity of a reinforcing steel bar point cloud which is weak relative to the number of concrete point cloud points is maintained. A preform point cloud cutting strategy is performed that cuts a single preform point cloud into a plurality of samples.
The point cloud cutting specifically comprises the following steps: and calculating the coordinate ranges of the prefabricated part point cloud coordinates after data enhancement in the x dimension, the y dimension and the z dimension, dividing the coordinates in the three dimensions according to a certain threshold range, and realizing the division of the prefabricated part point cloud after data enhancement based on the division result.
Different types of prefabricated parts have different ways of cutting the point cloud. As shown in fig. 6, the precast beam and the column point cloud are cut along the long axis thereof; for precast slabs, the wall point cloud is a line cut made by reducing a certain size along the edge of the largest concrete plane.
Point cloud downsampling: three-dimensional point clouds require down-sampling because the data is typically too large to compute. The traditional downsampling algorithm uniformly selects points in the grid, data are sparse and uniform, critical points are insufficient, and feature extraction is difficult. In the present invention, a curvature-based sampling method is used for down-sampling. The sampling algorithm has more sampling points in the area with more geometrical characteristics (larger curvature) of the point cloud, and the points on the surface of the steel bar or the joint of the steel bar and the concrete, which are considered as key area parts, can be reserved to the greatest extent, thereby being beneficial to feature extraction. The region division of the geometric features enables the sampling result to have strong noise resistance and high stability. Furthermore, the local distribution of the sampling points is uniform.
Fig. 6(a) - (d) are schematic diagrams of prefabricated columns, beams, plates and wall point clouds after point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling processing.
The prefabricated part point cloud data set with the marked information, which is manufactured by the method, is used for training by a PointNet + + component segmentation network, and the trained model tests the prefabricated part point cloud obtained by the three-dimensional scanner to obtain a better segmentation effect (the accuracy of the test set is 0.98289), so that the concrete part and the steel bar part of the prefabricated part are successfully identified. The data set generated by the method can replace a prefabricated part point cloud data set obtained by scanning to a certain extent and is used for practical application, and a new idea is provided for obtaining the prefabricated part point cloud data set in deep learning.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (8)
1. A prefabricated part point cloud data set manufacturing method for a deep learning segmentation network is characterized by comprising the following steps:
(1) determining the design sizes of concrete and steel bars in the prefabricated parts according to the design specifications of different types of prefabricated parts;
(2) calculating the area of a plane where the concrete and the steel bars can generate point clouds according to the design sizes of the concrete and the steel bars in the prefabricated part, wherein the area is in millimeter square as an area unit, introducing an eta parameter to express the integrity degree of the steel bar point clouds in the calculation process of the point number of the steel bar point clouds, and linking the area of the plane where the point clouds are generated with the point number of the point clouds which can be generated to obtain the point number of the point clouds which can be generated by the concrete and the steel bars;
(3) acquiring a coordinate matrix capable of generating the point cloud based on the geometrical characteristics of the plane of the point cloud generated by the concrete and the steel bars and the number of points capable of generating the point cloud, and generating the concrete point cloud and the steel bar point cloud according to the coordinate matrix capable of generating the point cloud and the Python environment;
(4) introducing density adjustment parameters to adjust the point densities of the concrete point cloud and the steel bar point cloud;
(5) after the point densities of the concrete point cloud and the reinforcing steel bar point cloud are adjusted, adding respective marking information to the concrete point cloud and the reinforcing steel bar point cloud to generate various reinforcing steel bars and concrete point clouds with marking information;
(6) combining the various steel bars with the marked information generated in the step (5) with the concrete point cloud to form various prefabricated part point clouds;
(7) and (4) sequentially carrying out point cloud normalization, point cloud data enhancement, point cloud cutting and point cloud down-sampling on the prefabricated part point cloud formed in the step (6) to obtain a prefabricated part point cloud data set with marked information.
2. The method for manufacturing a prefabricated part point cloud data set for a deep learning segmentation network according to claim 1, wherein the step (1) specifically comprises the following sub-steps:
(1.1) designing concrete a with different design sizes according to the design specifications of different types of prefabricated parts, wherein the concrete a is a cuboid, and the length, the width and the height of the concrete a are respectively l a 、w a 、h a The plane of the concrete a capable of generating the point cloud comprises five parts, namely a top plane, a left plane, a right plane, a front plane and a rear plane of a cuboid;
(1.2) designing steel bars with different design sizes according to the design specifications of different types of prefabricated parts, wherein the steel bars comprise straight steel bars, hook steel bars and U-shaped steel bars;
the design size of the straight steel bar is as follows: the straight reinforcing steel bar i consists of a cone and a cylinder, and the radiuses of the upper bottom surface and the lower bottom surface of the cone are r respectively i2 、r i1 Height of h i1 (ii) a Height of cylinder is h i2 Radius of the bottom surface is r i2 (ii) a The plane on which the straight steel bar i can generate the point cloud comprises three parts, namely a circular truncated cone side surface, a cylinder side surface and a cylinder upper bottom surface;
the design size of the hook steel bar is as follows: the hook steel bar j consists of a circular truncated cone, a first cylinder, a partial circular ring body and a second cylinder, and the radiuses of the upper bottom surface and the lower bottom surface of the circular truncated cone are r respectively j2 、r j1 Height of h j1 (ii) a The height of the first cylinder is h j2 Radius of the bottom surface is r j2 (ii) a The circular section of the partial circular ring body, i.e. the radius of the generating circle is r j2 The distance from the center of the generatrix to the axis of the ring is R j The central angle is alpha, and the alpha represents the bending angle of the steel bar of the hook steel bar; the radius of the bottom surface of the second cylinder is r j2 Height is l j (ii) a The plane on which the hook steel bar j can generate the point cloud comprises five parts, namely a circular truncated cone side surface, a first cylinder side surface, a partial circular ring side surface, a second cylinder side surface and a second cylinder upper bottom surface;
the design size of the U-shaped reinforcing steel bar is as follows: the U-shaped steel bar k consists of a first circular truncated cone body, a first cylinder, a half circular ring body, a second cylinder and a second circular truncated cone body, wherein the radiuses of the upper bottom surface and the lower bottom surface of the first circular truncated cone body are r respectively k2 、r k1 Height of h k1 (ii) a The height of the first cylinder is h k1 Radius of the bottom surface is r k2 (ii) a The radius of the circular section, i.e. the generatrix, of the half circular ring body is r k2 The distance from the center of the generatrix to the axis of the ring is R k The central angle is pi; the height of the second cylinder is h k1 Radius of the bottom surface is r k2 (ii) a The upper and lower bottom radii of the second circular truncated cone are r k2 、r k1 Height of h k1 (ii) a The plane of the U-shaped steel bar kj capable of generating the point cloud comprises five parts, namely a first circular truncated cone side surface, a first cylinder side surface, a half circular ring side surface, a second cylinder side surface and a second circular truncated cone side surface.
3. The method for producing the prefabricated part point cloud data set for the deep learning segmentation network according to claim 2, wherein the step (2) comprises the following sub-steps:
(2.1) calculating the area of the plane of the concrete capable of generating the point cloud, wherein the area unit is millimeter square, and relating the area with the number of points, wherein the number of the points of the concrete capable of generating the point cloud is equal to the area of the plane of the concrete capable of generating the point cloud; the calculation formula of the number of points of the concrete capable of generating the point cloud is as follows:
wherein a is the concrete type number; s 1 、S 2 、S 3 、S 4 、S 5 The area of the top plane, the area of the left side plane, the area of the right side plane, the area of the front side plane and the area of the rear side plane of the concrete a are respectively, and the square of a millimeter is taken as an area unit; s a The area of a plane capable of generating point clouds for the concrete a and the number of points of the point clouds correspondingly generated for the concrete a; s concrete For the number of all the points in the prefabricated part point cloud, the concrete with different design sizes in b is shared, n a The amount of concrete a;
(2.2) calculating the area of a plane of the steel bar capable of generating the point cloud, wherein the area unit is millimeter square, introducing an eta parameter to express the integrity degree of the steel bar point cloud, and then calculating the number of points of the steel bar capable of generating the point cloud; eta is obtained by dividing theta by 2 pi, and theta is a central angle corresponding to an arc formed by projecting an area with point cloud distribution on the side part of the cylinder in the steel bar to the bottom plane of the cylinder;
number of points S of steel bar point cloud in prefabricated part rebar Calculating the point number S of the point clouds of straight steel bars, hook steel bars and U-shaped steel bars straight 、S curved 、S U-shaped Adding to obtain;
a) number of points S of straight reinforcing steel bar capable of generating point cloud straight The calculation process of (2):
calculating the area of a plane of which the point cloud can be generated by the straight steel bars, wherein the unit of the area is millimeter square, and introducing eta i Parameters relate area to number of points, and straight steel bars can be generatedThe number of points of the point cloud is equal to the area of a plane in which the point cloud can be generated by the straight steel bars multiplied by eta i (ii) a The calculation formula of the point number of the point cloud generated by the straight steel bars is as follows:
wherein i is the type number of the straight steel bar; s. the iA 、S iB 、S iC The area of the side surface of the circular truncated cone, the area of the side surface of the cylinder and the area of the upper bottom surface of the cylinder of the straight steel bar i are respectively taken millimeter square as an area unit; s i Number of points, η, of point clouds generated for straight bars i i Representing the integrity degree of the steel bar point cloud generated by the straight steel bar i; s straight For the number of all points in the point cloud of the straight reinforcing steel bars in the prefabricated part point cloud, o kinds of straight reinforcing steel bars with different design sizes are provided, and n i The number of the straight steel bars i;
b) number of points S of hook steel bar capable of generating point cloud curved The calculation process of (2):
calculating the area of the plane of the hook steel bar capable of generating point cloud, wherein the unit of the area is millimeter square, and introducing eta j The parameters relate the area to the number of points, the number of points of the point cloud generated by the hook steel bar is equal to the product of the area of the plane of the point cloud generated by the hook steel bar and eta j (ii) a The calculation formula of the point number of the point cloud generated by the hook steel bar is as follows:
wherein j is the type number of the hook steel bar; s jD 、S jE 、S jF 、S jG 、S jH The area of the side surface of the circular truncated cone body of the hook steel bar j, the area of the side surface of the first cylinder body, the area of the side surface of the partial circular ring body, the area of the side surface of the second cylinder body and the area of the upper bottom surface of the second cylinder body are respectively in unit of millimeter square; s j Number of points, η, of point clouds correspondingly generated for the hook reinforcement j j Indicates that the hooked reinforcement j corresponds toThe integrity of the generated steel bar point cloud; s curved For the point number of all the hook steel bar point clouds in the prefabricated part point cloud, p hook steel bars with different design sizes are shared, and n j The number of the hooked steel bars j;
c) number of points S of point cloud can be generated by U-shaped steel bar U-shaped The calculation process of (2):
calculating the area of a plane of the point cloud generated by the U-shaped steel bar, wherein the unit of the area is millimeter square, and introducing a parameter eta k The area is related to the number of points, and the number of the points of the point cloud generated by the U-shaped steel bar is equal to the product of the area of a plane capable of generating the point cloud by the U-shaped steel bar and eta k (ii) a The calculation formula of the number of points of the point cloud generated by the U-shaped steel bar is as follows:
wherein k is the type number of the U-shaped steel bar; s kM 、S kN 、S kO 、S kP And S kQ The area of the side surface of the first circular truncated cone body, the area of the side surface of the first cylinder body, the area of the side surface of the half circular ring body, the area of the side surface of the second cylinder body and the area of the side surface of the second circular truncated cone body of the U-shaped steel bar k are respectively taken millimeter square as an area unit; s k Number of points, η, of point clouds correspondingly generated for the U-shaped reinforcement k k Representing the integrity degree of a reinforcing bar point cloud generated by the U-shaped reinforcing bar k; s U-shaped For the number of points of all the U-shaped steel bar point clouds in the prefabricated part point cloud, q U-shaped steel bars with different design sizes are arranged, and n k The number of the U-shaped reinforcing bars k.
4. The method for preparing a prefabricated part point cloud data set for a deep learning segmentation network according to claim 1, wherein the step (6) comprises the following sub-steps:
(6.1) selecting a certain amount and variety of steel bar point clouds and concrete point clouds required by the generated prefabricated part;
(6.2) performing space operations such as rotation, translation and the like on the reinforcing steel bar point cloud selected in the step (6.1) to enable the reinforcing steel bar point cloud and the concrete point cloud to be combined into a pattern of a required prefabricated part;
and (6.3) removing the point cloud on the concrete surface covered by the bottom of the steel bar to generate the point cloud of the prefabricated part.
5. The prefabricated part point cloud data set manufacturing method oriented to the deep learning segmentation network according to claim 1, wherein the point cloud normalization in the step (7) is specifically: and calculating the average value of the x, y and z coordinates of all points in the prefabricated part point cloud to be used as a centroid coordinate, then calculating the distance d from the point farthest from the centroid to the centroid, subtracting the centroid coordinate from the coordinates of all points, and dividing the centroid coordinate by d to obtain the normalized prefabricated part point cloud coordinate.
6. The method for manufacturing the prefabricated part point cloud data set for the deep learning segmentation network according to claim 5, wherein the point cloud data enhancement in the step (7) is specifically as follows: and (4) performing three transformations of rotation, dithering and translation on the normalized prefabricated part point cloud coordinate to obtain the prefabricated part point cloud coordinate after data enhancement.
7. The method for manufacturing the prefabricated part point cloud data set for the deep learning segmentation network according to claim 6, wherein the point cloud cutting in the step (7) is specifically as follows: and calculating the coordinate ranges of the prefabricated part point cloud coordinates after data enhancement in the x dimension, the y dimension and the z dimension, dividing the coordinates in the three dimensions according to a certain threshold range, and realizing the division of the prefabricated part point cloud after data enhancement based on the division result.
8. The method for manufacturing the prefabricated part point cloud data set for the deep learning segmentation network according to claim 7, wherein the point cloud downsampling in the step (7) is specifically as follows: and (4) performing down-sampling on the point cloud of the segmented prefabricated part by adopting a curvature-based sampling method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210721603.5A CN115100272B (en) | 2022-06-17 | 2022-06-17 | Prefabricated part point cloud data set manufacturing method for deep learning segmentation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210721603.5A CN115100272B (en) | 2022-06-17 | 2022-06-17 | Prefabricated part point cloud data set manufacturing method for deep learning segmentation network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115100272A true CN115100272A (en) | 2022-09-23 |
| CN115100272B CN115100272B (en) | 2023-09-22 |
Family
ID=83292535
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210721603.5A Active CN115100272B (en) | 2022-06-17 | 2022-06-17 | Prefabricated part point cloud data set manufacturing method for deep learning segmentation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115100272B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160232256A1 (en) * | 2015-02-06 | 2016-08-11 | Autodesk, Inc. | Model builder |
| CN110060338A (en) * | 2019-04-25 | 2019-07-26 | 重庆大学 | A kind of prefabricated components point cloud recognition methods based on BIM model |
| CN110363861A (en) * | 2019-07-14 | 2019-10-22 | 南京林业大学 | Field crop three-dimensional reconstruction method based on laser radar point cloud |
| CN113313834A (en) * | 2021-06-30 | 2021-08-27 | 北京大学深圳研究生院 | Point cloud generation method based on geometric interpretability |
| CN113487721A (en) * | 2021-06-18 | 2021-10-08 | 浙江大学 | Automatic prefabricated part identification method based on three-dimensional point cloud |
| CN113487722A (en) * | 2021-06-18 | 2021-10-08 | 浙江大学 | Automatic concrete member detection method based on three-dimensional laser scanning method |
| CN113868733A (en) * | 2021-09-01 | 2021-12-31 | 浙江大学 | Automatic existing structure BIM model generation method based on three-dimensional point cloud |
| CN114492619A (en) * | 2022-01-22 | 2022-05-13 | 电子科技大学 | Point cloud data set construction method and device based on statistics and concave-convex property |
-
2022
- 2022-06-17 CN CN202210721603.5A patent/CN115100272B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160232256A1 (en) * | 2015-02-06 | 2016-08-11 | Autodesk, Inc. | Model builder |
| CN110060338A (en) * | 2019-04-25 | 2019-07-26 | 重庆大学 | A kind of prefabricated components point cloud recognition methods based on BIM model |
| CN110363861A (en) * | 2019-07-14 | 2019-10-22 | 南京林业大学 | Field crop three-dimensional reconstruction method based on laser radar point cloud |
| CN113487721A (en) * | 2021-06-18 | 2021-10-08 | 浙江大学 | Automatic prefabricated part identification method based on three-dimensional point cloud |
| CN113487722A (en) * | 2021-06-18 | 2021-10-08 | 浙江大学 | Automatic concrete member detection method based on three-dimensional laser scanning method |
| CN113313834A (en) * | 2021-06-30 | 2021-08-27 | 北京大学深圳研究生院 | Point cloud generation method based on geometric interpretability |
| CN113868733A (en) * | 2021-09-01 | 2021-12-31 | 浙江大学 | Automatic existing structure BIM model generation method based on three-dimensional point cloud |
| CN114492619A (en) * | 2022-01-22 | 2022-05-13 | 电子科技大学 | Point cloud data set construction method and device based on statistics and concave-convex property |
Non-Patent Citations (2)
| Title |
|---|
| B, WEIJIAN ZHAO A , ET AL.: "Automated recognition and measurement based on three-dimensional point clouds to connect precast concrete components", 《AUTOMATION IN CONSTRUCTION》, pages 1 - 16 * |
| MIRZAEI, KAVEH , ET AL.: "3D point cloud data processing with machine learning for construction and infrastructure applications: A comprehensive review.", ADVANCED ENGINEERING INFORMATICS《》, pages 1 - 17 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115100272B (en) | 2023-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108846888B (en) | Automatic extraction method for fine size information of ancient wood building components | |
| CN109658431B (en) | Rock mass point cloud plane extraction method based on region growth | |
| CN105931291B (en) | A kind of complete dental modeling method of digitlization | |
| CN114119488B (en) | Intelligent detection method for dimension quality of prefabricated laminated slab for factory | |
| CN107122528A (en) | A kind of floor plan parametrization can edit modeling method again | |
| CN113487722B (en) | Automatic concrete member detection method based on three-dimensional laser scanning method | |
| CN107392954A (en) | A kind of gross error point elimination method based on sequence image | |
| CN114037706A (en) | Precast beam segment surface flatness detection method based on three-dimensional point cloud model | |
| Macher et al. | Semi-automatic segmentation and modelling from point clouds towards historical building information modelling | |
| CN115239620A (en) | Point cloud-based automatic identification method and system for welding points of reinforcing mesh | |
| CN113570722A (en) | Surrounding rock crack information extraction and integrity coefficient rapid determination method | |
| CN107037738B (en) | Material removal method based on STL model in geometry Simulation of NC Machining | |
| CN108053485B (en) | Potato three-dimensional modeling method based on contour image | |
| Ying et al. | Identification and reconstruction of concrete mesostructure based on deep learning in artificial intelligence | |
| CN115100272B (en) | Prefabricated part point cloud data set manufacturing method for deep learning segmentation network | |
| CN113868904A (en) | Method for automatically generating existing structure finite element model based on three-dimensional point cloud | |
| CN113971718A (en) | Method for performing Boolean operation on three-dimensional point cloud model | |
| CN113763529A (en) | Transformer substation modeling method based on three-dimensional scanning | |
| CN116881622A (en) | Multi-objective optimization method for space arrangement of three-dimensional laser scanning measuring station with complex structure | |
| CN107643048B (en) | Automatic extraction method of tree measuring factor based on point cloud data | |
| CN115937149A (en) | Wall surface local deviation automatic detection method based on triangular gridding | |
| CN113283495B (en) | Aggregate particle grading method and device | |
| CN110120058B (en) | Method for generating compact outer boundary by using high scatter points | |
| CN114170160A (en) | Steel bar classified scanning planning method based on geometric model | |
| CN110274919A (en) | Cut-parts decorative pattern measurement method of parameters based on Faster R-CNN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |