WO2022226588A1

WO2022226588A1 - Method and system for generating a model of a structure

Info

Publication number: WO2022226588A1
Application number: PCT/AU2022/050384
Authority: WO
Inventors: Thomas Caska; Rakesh Routhu; Ross Hennessy
Original assignee: Aerologix Group Pty Ltd
Priority date: 2021-04-28
Filing date: 2022-04-27
Publication date: 2022-11-03

Abstract

There is disclosed a method for generating a model of a structure from a point cloud data set of the structure. The method may include the steps of: extracting from the point cloud data set an extracted point cloud data set based on one or more spatial parameters associated with the shape of the structure; processing the extracted point cloud data set to provide a substantially isolated point cloud data set of the structure; fitting one or more predefined geometric models that represent features of the structure to one or more predefined portions of the isolated point cloud data set in which the features are expected to be present; and providing the model based on a combination of fitted ones of the one or more predefined geometric models. A related system and further methods are also disclosed.

Description

Method and System for Generating a Model of a Structure

Related Applications

[001] This application claims priority from Australian provisional patent application no. 2021901246 filed on 28 April 2021, the contents of which are incorporated by reference.

Technical Field

[002] The invention relates to a method and a system for generating a model of an object such as a structure. More specifically, the invention relates to a method and a system for generating a model of a structure in the form of a vertical structure such as a tower.

Background

[003] Photogrammetry and/or lidar data may be used to generate point cloud data sets of objects such as structures and buildings. Such a point cloud data set may comprise many thousands of data points having X, Y, Z coordinates in space that may be in a relative or absolute frame.

[004] It is desirable to create CAD (Computer Aided Design) models from point cloud data such as models of structures and objects that may include, but not limited to, buildings, bridges, towers and roads or the like.

[005] To create a CAD model from point cloud data one or more model features such as a line, surface or shape needs to be fitted to the point cloud data. This is sometimes referred to as the vectorisation of the data. One method to do this is by presenting the point cloud data via a user interface (UI), and a user adding or drawing nodes associated with a selected model onto the point cloud data. The selected model is then applied between the nodes. [006] An example of a model feature may be the fitting of a polygon surface to a feature of interest, such as a road, by the user identifying at least 3 points on the road and then applying a polygon function to create the polygon model feature that is associated with the road.

[007] A problem with this method relates to a user needing to review the point cloud data using the UI and then manually applying the model features to the features of interest. Such a process may be slow, inaccurate and have poor repeatability. Another problem relates to the entire data set being present for processing and requiring substantive computing resources such as needing to use a UI to present all data to the user.

[008] The invention disclosed herein seeks to overcome one or more of the above identified problems or at least provide a useful alternative.

Summary

[009] In accordance with a first broad aspect there is provided, a method for generating a model of a structure from a point cloud data set of the structure, the method including the steps of: Extracting from the point cloud data set an extracted point cloud data set based on one or more spatial parameters associated with the shape of the structure; Processing the extracted point cloud data set to provide a substantially isolated point cloud data set of the structure; Fitting one or more predefined geometric models that represent features of the structure to one or more predefined portions of the isolated point cloud data set in which the features are expected to be present; and Providing the model based on a combination of fitted ones of the one or more predefined geometric models.

[0010] In an aspect, the one or more spatial parameters include predefined parameters.

[0011] In another aspect, the predefined parameters include at least one of a height, a width and a depth of the structure. [0012] In yet another aspect, the predefined parameters include a height, a width and a depth so as to define a 3 dimensional bound about the structure for the extracted point cloud data set.

[0013] In yet another aspect, the method includes identifying at least one of a top and a base of the point cloud data set, and defining the dimensional bound relative to at least one of the top and base.

[0014] In yet another aspect, the processing of the extracted point cloud data set to provide the substantially isolated point cloud data set of the structure includes performing data clustering.

[0015] In yet another aspect, the data clustering is region growing clustering.

[0016] In yet another aspect, the one or more predefined portions of the isolated point cloud data set in which the features are expected to be present are identified based on one or more predefined relative geometric positions associated with the structure within the isolated point cloud data set.

[0017] In yet another aspect, the predefined geometric models are preselected to fit shapes at or proximate the one or more predefined portions.

[0018] In yet another aspect, the predefined geometric models include one or more of cylinders, frusto-conical shapes and rectangular members.

[0019] In yet another aspect, the fitting of the predefined geometric models to one or more predefined portions of the isolated point cloud data set in which the features are expected to be present identifies and classifies associated isolated point cloud data as being representative of the fitted feature.

[0020] In yet another aspect, the identified and classified associated isolated point cloud data being representative of the fitted feature is at least temporary removed from the isolated point cloud data to facilitate efficient processing. [0021] In yet another aspect, the structure is a vertically projecting structure.

[0022] In yet another aspect, the structure is a tower standing on a ground surface.

[0023] In yet another aspect, the structure is a telecommunication tower with a pole supporting an antenna arrangement.

[0024] In yet another aspect, the one or more spatial parameters associated with the shape of the structure includes a spatial parameter representative of an estimated height of the tower.

[0025] In yet another aspect, the one or more predefined geometric models that represent features of the structure includes a pole model and an antenna model.

[0026] In yet another aspect, the model includes vectorised datapoints suitable for a CAD (Computer Aided Design) model.

[0027] In accordance with a second broad aspect there is provided, a system for generating a model of a structure from a point cloud data set of the structure, the system being configurable by software to: Extract from the point cloud data set an extracted point cloud data set based on one or more spatial parameters associated with the shape of the structure; Process the extracted point cloud data set to provide a substantially isolated point cloud data set of the structure; Fit one or more predefined geometric models that represent features of the structure to one or more predefined portions of the isolated point cloud data set in which the features are expected to be present; and Provide the model based on a combination of fitted ones of the one or more predefined geometric models.

[0028] In accordance with a third broad aspect there is provided, a method for generating a model of a structure from a point cloud data set of the structure, the structure including a support feature and a peripheral feature carried by the support feature, the method including the steps of: Extracting from the point cloud data set an extracted point cloud data set based on one or more spatial parameters associated with the shape of the structure; Processing the extracted point cloud data set to provide a substantially isolated point cloud data set of the structure; Fitting a first predefined geometric model that represents the support feature to a portion of the isolated point cloud data set in which the support feature is expected to be present and associating a first fitted portion of the isolated point cloud data set with the first predefined geometric model; Further processing the substantially isolated point cloud data set of the structure to at least temporarily remove from processing the first fitted portion of the isolated point cloud data set to provide a reduced isolated point cloud data set; Fitting a second predefined geometric model that represents the peripheral feature to a further portion of the reduced isolated point cloud data set in which the peripheral feature is expected to be present and associating a second fitted portion of the isolated point cloud data set with the fitted second predefined geometric model; Provide the model based on a combination of fitted ones of the first and second predefined models.

[0029] In accordance with a fourth broad aspect there is provided, a method for processing a point cloud data set of a structure to provide a model of the structure, the structure including a support feature and a peripheral feature carried by the support feature, the method including the steps of: Locating the structure within the point cloud data set and determining geometric bounds about the structure; Extracting from the point cloud data set an extracted point cloud data set within the geometric bounds; Fitting a first predefined geometric model that represents the support feature to a portion of the isolated point cloud data set in which the support feature is expected to be present and associating a first fitted portion of the isolated point cloud data set with the first predefined geometric model; Processing the substantially isolated point cloud data set of the structure to provide a reduced isolated point cloud data set in which the first fitted portion of the isolated point cloud data set is removed; Fitting a second predefined geometric model that represents the peripheral feature to a further portion of the reduced isolated point cloud data set in which the peripheral feature is expected to be present and associating a second fitted portion of the isolated point cloud data set with the fitted second predefined geometric model; Providing the model based on a combination of fitted ones of the first and second predefined models.

[0030] In accordance with a fifth broad aspect there is provided, a method for processing a point cloud data set of an object to provide a model of the object, the object including a hierarchy of features of varying geometric complexity, the method including the steps of: Locating the object within the point cloud data set and determining geometric bounds about the object; Extracting from the point cloud data set an extracted point cloud data set within the geometric bounds; Fitting a first predefined geometric model that represents at least a first of the hierarchy of features to a portion of the isolated point cloud data set in which the first of the hierarchy of features is expected to be present and associating a first fitted portion of the isolated point cloud data set with the first predefined geometric model; Processing the substantially isolated point cloud data set of the object to provide a reduced isolated point cloud data set in which the first fitted portion of the isolated point cloud data set is removed; Fitting a second predefined geometric model that represents at least a second of the hierarchy of features to a further portion of the reduced isolated point cloud data set in which the second of the hierarchy of features is expected to be present and associating a second fitted portion of the isolated point cloud data set with the fitted second predefined geometric model; Providing the model based on a combination of fitted ones of the first and second predefined models.

[0031] In accordance with a sixth broad aspect there is provided, a method for sequentially processing a point cloud data set of an object to provide a computer aided design (CAD) model of the object using a set of predetermined models to represent at least some of a hierarchy of features associated with the object, the method including: sequentially fitting each of the predetermined models to the point cloud data and upon fitting of each of the predetermined models, removing data associated with fitted ones of the predetermined models to provide sequentially reduced point cloud data for processing and fitting of further ones of the predetermined models.

[0032] In accordance with a seventh broad aspect there is provided, a method for sequentially processing a point cloud data set representing an object to provide a computer aided design (CAD) model of the object using a set of predetermined models to represent at least some of a hierarchy of features associated with the object, the method including: sequentially identifying point cloud data expected to be associated with each of the hierarchy of features, fitting each of the predetermined models to the identified point cloud data and upon fitting of each of the predetermined models, removing data associated with fitted ones of the predetermined models to provide sequentially reduced point cloud data for identification of further ones of the hierarchy of features and fitting of further ones of the predetermined models thereto.

Brief Description of the Figures

[0033] The invention is described, by way of non-limiting example only, by reference to the accompanying figures, in which;

[0034] Figure l is a block diagram illustrating a system for the generation of a model of a structure;

[0035] Figure 2 is a flow diagram illustrating a first example method for the generation of a model of a structure;

[0036] Figure 3 is a flow diagram illustrating a second example method of the generation of a model of a structure;

[0037] Figure 4 is a flow diagram illustrating a third example method of the generation of a model of a structure, in the form of a monopole tower structure;

[0038] Figures 5a to 5p illustrate example data input and outputs of the first and second example methods as well as providing a more detailed third example of a method the generation of a model of a structure focussed on a monopole telecommunication tower. The datapoints are shown as coloured visual representations for ease of understanding. A more detailed explanation of each Figure is provided below;

[0039] Figure 5a is a visualisation of point cloud data illustrating an extracted portion of data associated with, in this example, an upper portion of a structure;

[0040] Figure 5b is a visualisation of point cloud data illustrating an extracted portion of data associated with a geometric bound projected downwardly to obtain data points in the region of the structure;

[0041] Figure 5c is a visualisation of point cloud data illustrating a ground surface to which a plane may be fixed;

[0042] Figure 5d is a visualisation of point cloud data illustrating features below the ground surface being removed;

[0043] Figure 5e is a visualisation of data illustrating region growing clustering being applied to the point cloud data;

[0044] Figure 5f is a visualisation of point cloud data after the region growing clustering has been performed illustrating the isolated point cloud data of the structure;

[0045] Figure 6a is a visualisation of data illustrating a model fitted to part, in this example the base, of the structure;

[0046] Figure 6b is a visualisation of data illustrating a portion of the point cloud data above the base;

[0047] Figure 6c is a visualisation of data illustrating a main portion of the structure, in this case the monopole, being fitted with a model to represent the main portion;

[0048] Figure 6d is a visualisation of data illustrating an extension of the model upwardly to a top of the monopole;

[0049] Figure 6e is a visualisation of data illustrating an example of a resultant three- dimensional CAD model of the structure including its base and main monopole;

[0050] Figure 7a is a visualisation of data illustrating an extracted portion of the data being processed to determine a curvature of features of the periphery features of the structure, in this example, antennas;

[0051] Figure 7b is a visualisation of data illustrating the application of region growing clustering based on the curvature to identify candidate periphery features, in this example, the antennas; [0052] Figure 7c is a visualisation of data illustrating processing to identify geometric features, in this example planes associated with front faces, of the antennas and identify the antennas;

[0053] Figure 7d is a visualisation of data illustrating processing to identify orientations of the identified antennas;

[0054] Figure 7e is a visualisation of point cloud data illustrating antennas relative to an antenna cluster at a top of the structure;

[0055] Figure 7f is a visualisation of data illustrating identified antennas being passed through a clustering algorithm to obtain a set of antenna models;

[0056] Figure 7g is a visualisation of data illustrating identified antennas being shown relative to the point cloud dataset;

[0057] Figure 7h is a visualisation of data illustrating the antennas that have been fitted and classified into a group, each colour representing a different one of six identified groups;

[0058] Figure 71 is a visualisation of data illustrating the identified antenna groups relative to the remaining structure;

[0059] Figure 7J is a visualisation of the point cloud data illustrating identified further vertical structures, shown in red, after the main pole features and identified antennas have been removed from the dataset;

[0060] Figure 7K is a visualisation of the point cloud data illustrating vertical structing being removed to reveal torus or ring like shapes;

[0061] Figure 7L is a visualisation of data illustrating matching torus or ring like models to the identified torus or ring like shapes;

[0062] Figure 7M is a visualisation of data illustrating matched torus or ring like models located relative to the extracted point cloud data;

[0063] Figure 7N is a visualisation of data illustrating geometric surface models fitted to the torus or ring like shapes and surfaces;

[0064] Figure 70 is a visualisation of data illustrating further outer peripheral support structures being modelled using, in this example, half torus structures;

[0065] Figure 7P is a visualisation of data illustrating output CAD model of the peripheral support, in this example, the upper monopole tower support; and

[0066] Figure 8 is a visualisation of data illustrating output CAD model of a further example of a monopole telecommunication tower generated by the method disclosed herein.

Detailed Description

System and Introduction

[0067] Referring to Figure 1 there is shown a system 10 for the processing of data to provide a model 12 of an object such as a structure 14, examples of which are shown in Figures 4a to 6f. The system 10 may include one or more computer systems 11 that are configured to processed data received from a data acquisition means 16. The data acquisition means 16 may include one or a combination of an image capturing device and sensors such as, but not limited to, lidar devices. The data captured by the data acquisition means 16 is either provided as point cloud data or is processed by the one or more computer systems 11 to provide point cloud data. For example, lidar signals may be processed to provide (C,U,Z) point cloud data and digital photographic data may be processed using photogrammetry techniques to provide similar (C,U,Z) point cloud data. Both examples of contemplated herein.

[0068] The data acquisition means 16 may be used to collect data from and about the object or structure 14 such as, but not limited to, a man-made structure such as a bridge, road, tower or the like, or a natural structure such as a river or rock formation. Although, the system and methods described herein find preferable use with man-made structures, in particular, structures that have some identifiable geometric shapes (i.e cylinders, rectangles etc) such as vertical structures like towers. The data acquisition means 16 may be carried by a vehicle, manned or unmanned, such as a drone. An aerial drone being the most suitable vehicle to collect data about vertical structures using one or both of lidar and photogrammetry techniques.

[0069] The system 10, namely, the one or more computer systems 11 are configurable by software to operate the methods described herein. The one or more computer systems 11 include respective processors 18, memory 20 and may include a database 22. The one or more computer systems 11 may include networks servers that provide cloud- based services to user terminals or may be stand-alone computing devices. All such configurations are considered herein.

[0070] The system 10 is ultimately configured to process the collected point cloud data of the structure into vectorized date for a CAD (Computer Aided Design) suitable model that represents the structure. Such a generated CAD model finds use to, for example, perform engineering tasks such as maintenance planning, upgrades and confirming plans and dimensions.

[0071] Structures such as, for example, telecommunication towers or bridges can be complex structures having both core structural features such as supports and peripheral features such as antennas, flag poles or the like. As such, collected point cloud data may include many thousands, if not millions, of data points to capture all of these features. Further, the collected point cloud data may include data from obstructions such as trees or artifacts that are not part of the structure required to be modelled. As such, the processing of collected point cloud data is computationally expensive, and in some instances, not computable because the data is simply too complex and any forms or shapes not identifiable.

[0072] To address this problem, methods disclosed herein seek to use domain constraints and domain specific knowledge to process the data, and make the processing of the data more efficient and accurate. For example, when a structure is being modelled - some knowledge of the structure may be used to assist the data processing such as the expected height of the structure, any identified georeferenced information, and the expected shape of the structure. For example, with a monopole structure is it expected that the main support of the structure is generally cylindrical and extends from a ground surface. This knowledge can be used to predefine geometric shapes to be fitted to specific portions of the point cloud data. In some examples, each type of structure such as a tower or bridge, may have a catalogue of predefined geometric shapes to fit with that structure. The object or structure may include or be broken into a hierarchy of features ranging from simpler more readily identifiable features and structures to more complex features and structures. The simpler features and structures may be processed firstly.

[0073] Further, once fitted, the point cloud data already associated with the predefined geometric shapes can be removed or tagged to be omitted from processing to focus further processing. For example, main supports may be processed firstly leaving behind the data associated with more complex peripheries for further processing. Such a sequential method of processing, that may include the hierarchy of processing preferably starting from the main or simpler known structures or features firstly and at least temporary removal of this data, allows the data set being processed to be reduced during processing thereby reducing the computational expense. In other words, the point cloud data is sequentially reduced through the processing which makes it easier to identify and process features remaining in the data set, especially more complex features.

[0074] The model of the structure may be formed by combining the fitted geometric shapes that may be identified as, for example, main pole, brackets, antenna, beams or the like and this allows a model parts catalogue to be created and save to, for example, the database.

First Example Method

[0075] Turning to a first method 100 for generating a model of a structure from a point cloud data set of the structure, the method 100 may include: at step 110, extracting from the point cloud data set an extracted point cloud data set based on one or more spatial parameters associated with the shape of the structure. For example, a top of the structure may be identified, and an approximate height and width of the structure is known. As such, a [Z] coordinate that defines the top and an [X,Y] plane that defines the overall projected width may be determined and projected downwardly to define a portion of the point cloud data that forms the extracted point cloud data set. A ground surface may also be determined to define the bottom of the structure. This allows data points below and/or at the ground surface to be removed.

[0076] At step 120, the method 100 includes processing the extracted point cloud data set to provide a substantially isolated point cloud data set of the structure. For example, data clustering such as region growing data clustering may be used to filter out the main tower and thereby provide the isolated point cloud data set of the structure. The isolated point cloud data set now includes only the data points of the structure.

[0077] At step 130, the method 100 includes fitting one or more predefined geometric models that represent features of the structure to one or more predefined portions of the isolated point cloud data set in which the features are expected to be present. For example, the system 10 may store a preconfigured set of predefined geometric models associated with the structure that is being modelled. For example, a monopole structure may include a cylindrical or canonical frustum to model its main support pole portion and may include a plurality of rectangular based models that may be fitted to antenna as is further described below.

[0078] As further described in the more specific examples below, the larger better- defined features may be identified from an expected geometric position such as the main pole structure extending from the base and may be modelled before the more complex periphery structures. Then, once such a base or support model is created and fitted, its associated data points may be removed or at least tagged for no further processing so that other unmodeled datapoints may be modelled.

[0079] At step 140, the method includes providing the model based on a combination of fitted ones of the one or more predefined geometric models. For example, the cylindrical or canonical frustum to model may be provided, and the plurality of rectangular models displayed thereabout to represent the antennas. A similar method may be applied to a variety of structures and is not limited to tower or monopole-based stmctures. The model may be CAD model in a format suitable to be read by a CAD program.

Second Example Method

[0080] Turning to a second example method 200 for generating a model of a structure from a point cloud data set of the structure. The structure 14 may include one or more support or main features and one or more peripheral features. The support features may include, but not limited to, a beam or a pole in a variety of orientations such as vertical, angled and horizontal. The peripheral features may be more complex features and may include, but not limited to, smaller appendages such antenna, scaffolding, cables or the like. In the example shown in Figure 7b, the support may include the main vertical pole of the monopole and the appendages may include the antenna array located toward to top pf the monopole.

[0081] At step 210, similarly to method 100, the method 200 includes extracting from the point cloud data set an extracted point cloud data set based on one or more spatial parameters associated with the shape of the structure. For example, a top of the structure may be identified, and an approximate height and width of the structure are known. As such, a [Z] coordinate that defines the top and an [X,Y] plane that defines the overall projected width may be determined and projected downwardly to define a portion of the point cloud data that forms the extracted point cloud data set. A ground surface may also be determined to define the bottom of the structure. This allows data points below and/or at the ground surface to be removed.

[0082] At step 220, again similarly to method 100, the method 200 includes processing the extracted point cloud data set to provide a substantially isolated point cloud data set of the structure. For example, data clustering such as region growing data clustering may be used to filter out the main tower and thereby provide the isolated point cloud data set of the structure. The isolated point cloud data set now includes only the data points of the structure.

[0083] At step 230, the method 200 includes fitting a first predefined geometric model that represents the support feature to a portion of the isolated point cloud data set in which the support feature is expected to be present and associating a first fitted portion of the isolated point cloud data set with the first predefined geometric model. The fitting of the first predefined geometric model to the support feature may include, but not limited to, fitting a cylindrical, rectangular or frustum to a main support. In the example of a monopole this may include, as is further identified below, locating a base plane of the tower and then fitting a cylinder and frustum to model the base portion and then elongate upper portion of the monopole. The system 10 may be preloaded with a range of canonical models and these may be automatically selected and applied based on the domain knowledge, more specifically the geometric regions where the features are expected to be present.

[0084] At step 240, the substantially isolated point cloud data set of the structure is then further processes to at least temporality remove or at least not process the first fitted portion of the isolated point cloud data set to provide a reduced isolated point cloud data set. Accordingly, data points associated with already modelled features are removed or tagged not for further processing. This step may be repeated for other support type features to provide the main features of a structure.

[0085] At step 250, the method 200 includes fitting a second predefined geometric model that represents the peripheral feature to a further portion of the reduced isolated point cloud data set in which the peripheral feature is expected to be present and associating a second fitted portion of the isolated point cloud data set with the fitted second predefined geometric model. For example, the further portion may include an upper section of the dataset toward the top of structure where peripheral features such as antennas may be present. The second predefined geometric model may be selected from predefined models representative of the peripheral feature such as, but not limited to, antennas. This step may be repeated for all peripheral type features.

[0086] At step 260, the method 200 includes generating the model based on a combination of fitted ones of the first and second predefined models. For example, the cylindrical or canonical frustum model may be provided, and the plurality of rectangular based models displayed thereabout to represent the antennas. A similar method may be applied to a variety of structures, and it not limited to tower or monopole-based structures. The model may be a CAD model in a format suitable to be read by a CAD program.

Third Example Method

[0087] Referring now to Figure 4 there is shown more detailed third example of a method 300 for the generation of the model 12 of the structure 14 in the example form of a monopole telecommunication tower 50.

[0088] The data associated with the method steps are shown visually in Figures 5a to 7p for ease of explanation that may be data or outputs associated with the first and second example methods 100, 200 as well as providing a more detailed third example of the method 300 for the generation of the model 12 of the structure 14 in the example form of a monopole telecommunication tower 50.

[0089] In more detail, Figures 5a to 5f illustrate method steps and an output to locate and isolate the point cloud data of the tower 50. Figures 6a to 6e illustrate the method steps and outputs of fitting a geometric model to the pole 54 of the tower and the resulting model of a base 52 and the pole 54 of the tower 50. Figures 7a to 7f illustrate method steps and outputs of fitting a geometric model to more complex periphery features of the tower 50 such as antennas 56 and related antenna support structures 58. An example of a completed tower model 12 is shown in Figure 8.

[0090] Turning firstly to Figures 5a to 5f, the steps of locating the structure in the form of the tower 50 from a point cloud data set 40 are shown. At step 310, the method 300 includes extracting a top portion 42 of the point cloud data set based on domain knowledge. In this example, the domain knowledge is that the tower is the most likely the only data at the highest Z interval in the data set and the top portion of the point cloud data may include, but not limited to, the top 10% of the data points. This could be improved further by clustering these top points and removing “outliers” by finding the “highest density” cluster towards the centre of the data. The output of this step when run on the example tower model data is acceptable without the further filtering as shown in Figure 6a. [0091] At step 320 as shown in Figure 5b an (X,Y) bounds based on the top portion are projected downwardly through the entire point cloud data set to get all points in the region of the tower 50. This could be modified if needed by “growing” the cluster in a negative z direction if the top of tower (X,Y) bounds wasn’t the largest extent of the whole tower.

[0092] At step 330 as shown in Figure 5c, an (X,Y) plane may be fitted or fixed to find the ground surface 62. Identifying of the ground plane is used to estimate the tower Z vector from (assuming the local coordinate frame isn’t perfectly aligned with the tower). This may involve the examining the bottom 10% of points to fit an (X,Y) plane representing the ground surface. To perform this segmentation a RANSAC (Random Sample Consensus) algorithm is used to search for a plane with the normal vector with 5 degrees of the Z unit vector (0,0,1). This works well on the example dataset, but is another area which could be optimised if required for data where the mounting surface of the tower may not be very level, as shown in Figure 5c.

[0093] At step 340 as shown in Figure 5d, data points from the base 52 of the tower 50 and below are removed. After finding the ground surface plane, the model is used to filter out the points which lie within close proximity to this plane. The dataset may then be filtered down to remove these points, which leaves only the tower remaining as shown in Figure 5d.

[0094] At step 350 as shown in Figure 5e, the remaining dataset now should represent only the tower 50 to be modelled, however it can see from the previous Figure that it also includes surrounding objects considered as “noise” data points associated with the tower are filtered out using a region growing cluster algorithm which at step 360 as shown in Figure 5f an extracted point cloud data set 70 with the tower features being isolated is provided. Models may then be fitted to this data as is further described below.

[0095] Turning now to Figures 6a to 6e, there is shown the generation of a model of the support features of the tower, in this example, being the base 52 and main monopole 54. At step 370 as shown in Figure 6a, a predetermined geometric shape, in this case a cylinder, is fitted to the base 52 of the tower 50 using a RANSAC algorithm to get points which best model a constrained cylinder which leaves the main pole 54 above the base 52 at step 380 as shown in Figure 6b. In more detail, the method may include searching using a RANSAC algorithm for a cylinder with a diameter in the range of 20cm up to 1.5m. This may be performed over small segments to “z” height intervals, then filtered and analysed to get the exact height of the base cylinder.

[0096] At step 390 as shown in Figure 6c, a further support feature, being the main pole 54, is modelled using a predetermined geometric shape, in this case a canonical frustrum, again with an iterative RANSAC algorithm. In more detail, the method includes iteratively analysing small sections of “z” height of the data to determine the radius of the pole, then performs a least-squares polynomial fit to this array of radius estimates to determine the effective opening angle for the canonical frustrum model to be fitted or more accurately the constraints on the range of values that will be attempted to be fitted.

[0097] After fitting the model, a convex hull crop may be performed to accommodate for all gaps in the data where support structures are attached to the pole at step 400 the model is extended upwardly to the remaining pole points as shown in Figure 6d. In more detail, this model also matches points right up through where there are a lot of occlusions on the tower pole due to the surrounding antennas and attached cabling. The method extends up through to the top and then perform some filtering based on the histogram of Z values to accurately determine the tope of pole.

[0098] As shown in Figure 6e, at step 410 a 3-dimensional CAD model of the support features of the tower 50 is provided including a base plane (ground surface) 62, base 52 and frustoconical monopole 54.

[0099] Figures 7a to 7p illustrate further method steps relating to the identification and modelling of periphery features of the tower 50 which are in this example the tower antennas 56 and support members 58 for the antennas 56. It is noted that the substantially isolated point cloud data set of the structure may be processed to provide a reduced isolated point cloud data set in which the already modelled data points, such as those associated with the tower base and main monopole are removed from the isolated point cloud data set. Further, domain knowledge may be introduced to select a portion of the data set for processing. For example, in this case, the upper portion of the tower includes the peripheral features and therefore only the upper portion of the reduced isolated point cloud data set is required to be processed.

[00100] At step 420 as shown in Figure 7a, this stage of the modelling, the method seeks to classify the antennas 56 in the dataset. This part of the method may begin by calculating the curvature of the remaining antenna dataset (with the pole model points removed) using a spatial tolerance and a representation of the curvature, indicated by colour, of the data set is provided in Figure 7a.

[00101] Then, at step 430 as shown in Figure 7b, a region growing clustering algorithm based on the curvature is used to find candidate antenna clusters which are shown in Figure 7c. In more detail, for output shown in Figure 7b, the following parameters were used (min_cluster_size=20000, neighbours=100, smoothness_deg_threshold=5, curvature_threshold=0.05) to refine these antenna clusters to filter down to just the antenna (trimming off the support structure and cabling). The method started with a PCA (Principal Component Analysis) calculation, to estimate a more accurate z vector for the mounted antenna than unit z.

[00102] Then the method filters out points from the top and bottom of the antenna (mostly brackets and cables) using a hull area ratio threshold. The ratio for top and bottom are tuned separately and is an area for future improvements depending on the structure of the dataset. At step 440, RANSAC is used to match planes and look for the front face of the antenna and a step 450, as shown in Figure 7d - an axis “A” may be defined for each matched antenna 56 to define its position and orientation. The RANSAC algorithm matches back planes close to the estimated z vector normal and project through to the front face of the antenna.

[00103] At step 460, once the candidate antenna points are modelled, the type of antenna needs to be estimated, an example of multiple antenna modelling can be seen in the point cloud data set shown in Figure 7e.

[00104] At step 470 as shown in Figure 7f, each candidate antenna pointset can be normalised and may be pass through a clustering algorithm to get the set of antenna models present in the dataset. These may be extracted and displayed which in this case include six different types, indicated as 56a to 56f. In more detail, this method involves effectively normalising each candidate antenna pointset into its own local frame, while also down sampling using an octree voxel to ensure a consistent sampled density. Then, the method involves passing all normalised antenna datasets through a clustering algorithm to get the set of antenna models present in the dataset. This clustering algorithm uses ICP (Iterative Clustering Projection) to compare each antenna dataset against every other to generate a score for the match between the two. It then clusters the scores, effectively finding all groupings of different antenna models present in the tower dataset.

[00105] At step 480, after estimating the location of each antenna and candidate points on the antenna, the antenna models may be displayed over the dataset as shown in Figure 7g. At step 490 as shown in Figure 7h, each individual candidate antenna point cloud may be passed through an iterative algorithm which calculates a “fit” score for every antenna in the catalogue. Each antenna is then classified as the best fit model using a classifier. This gives us the six distinct antenna model groups (56a to 56f), which can be seen according to the rendered point cloud colour in Figure 7h. At step 500 as shown in Figure 7i, once the antenna models are classified, the models can be exported to the combined CAD model. Point cloud data associated with modelled features may be removed from the dataset leaving behind more specific features as are further discussed below.

[00106] At step 510, as shown in Figure 7j, peripheral features in the form of antenna support structures 58 shown which include vertical structural supports 80 that may be extracted from the remaining reduced and isolated point cloud structure. In more detail, once the tower pole 54, and all antennas 56 are modelled, then the method seeks to model the support structure 58 for the antenna 56. The first step in this part of the method is to filter out the cabling support structure 82, and this may be accomplished by a Region Growing clustering algorithm, which looks for the largest “vertical” cluster as this support runs up the full distance of the pole 54. Then performing region growing clustering with low smoothness and curvature thresholds to identify fairly flat surfaces and subtract these to identify the points with relatively high curvature (e.g. The tubular structure). Then, modelling all vertical support pole structures 80 by using the cylinder RANSAC algorithm. [00107] At step 520, the extracted vertical structural supports 80 may be used to refine the search space further peripheral features, in this case, torus shaped supports 84 as shown in Figure 7k. In more detail, this includes subtracting the matched vertical points and can also use the positions of these vertical structures to refine the search space for the torus shaped candidate datasets. That is, clusters of points which are radiating out from the tower in tower-aligned cylindrical coordinates.

[00108] At step 530 as shown in Figure 7L, the torus geometry is matched which results in step 540 with a joint support model shown in Figure 7m. This includes estimating the torus geometry models by transforming points into a local cylindrical coordinate space to get radius values, then applying peak extraction on a histogram of these radial values to identify clusters based on frequency.

[00109] At step 550, as shown in Figure 7n, an iterative radial search is performed inside the modelled structures to look for mesh panels to be modelled as surface geometry. This involves matching planar horizontal support surfaces 86, coplanar with the torus that has been matched. This can include one large surface within the matched torus radius, as well as multiple smaller surfaces outside of that radius. This is done by using the tori radius to constrain the search space (eg. Crop points within tori) and then applying a RANSAC algorithm to match an X,Y plane.

[00110] At step 560 outer horizontal support structures 86 and half torus 88 are matched as shown in Figure 7o. This includes matching the outer horizontal support structures by looking for slices of data which are coplanar with the matched tori. Then apply a RANS AC/MS AC with a d-section model defined by a partial torus and two attached cylinders. A possible improvement may be to grow the surface geometry matched in the previous stages to fill the modelled tori. The final model of the upper tower support geometry is provided at step 570 as shown in Figure 7p. The modelled features may be included in the combined 3 -dimensional CAD model 12 at step 580, an example of which is shown in Figure 8.

[00111] It is envisioned that the above system may in some examples additionally include a graphic user interface (GUI) that allows a user to allocate unmodelled data points to additional features such as cables or other complex features not accounted for in the predefined set of geometric models. Such as GUI could be used to fit further geometric features such as line and surfaces to these complex features and then added to the above-described models.

[00112] Advantageously, there has been described a method and a system for generating a model of a structure that uses domain knowledge and constraints to allow the efficient and automated processing of point cloud data to provide the model. The domain knowledge and constraints allow the data to be selectively processed for features that are expected to be present enabling automated model fitting. Further, data associated with modelled features may be sequentially removed, or otherwise not processed, from the remaining data being processed which reduces the computational cost and improves the ability to identify remaining features in the reduced dataset. As such, the method and system provide substantive technical advantages both in terms of model generation and within the computational implementation.

[00113] Throughout this specification and the claims which follow, unless the context requires otherwise, the word "comprise", and variations such as "comprises" and "comprising", will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

[00114] The reference in this specification to any known matter or any prior publication is not, and should not be taken to be, an acknowledgment or admission or suggestion that the known matter or prior art publication forms part of the common general knowledge in the field to which this specification relates.

[00115] While specific examples of the invention have been described, it will be understood that the invention extends to alternative combinations of the features disclosed or evident from the disclosure provided herein.

[00116] Many and various modifications will be apparent to those skilled in the art without departing from the scope of the invention disclosed or evident from the disclosure provided herein.

Claims

The claims defining the Invention are as follows:

1. A method for generating a model of a structure from a point cloud data set of the structure, the method including the steps of:

Extracting from the point cloud data set an extracted point cloud data set based on one or more spatial parameters associated with the shape of the structure;

Processing the extracted point cloud data set to provide a substantially isolated point cloud data set of the structure;

Fitting one or more predefined geometric models that represent features of the structure to one or more predefined portions of the isolated point cloud data set in which the features are expected to be present; and

Providing the model based on a combination of fitted ones of the one or more predefined geometric models.

2. The method according to claim 1, wherein the one or more spatial parameters include predefined parameters.

3. The method according to claim 2, wherein the predefined parameters include at least one of a height, a width and a depth of the structure.

4. The method according to claim 2, wherein the predefined parameters include a height, a width and a depth so as to define a 3 dimensional bound about the structure for the extracted point cloud data set.

5. The method according to claim 3, wherein the method includes identifying at least one of a top and a base of the point cloud data set, and defining the dimensional bound relative to at least one of the top and base.

6. The method according to claim 1, wherein the processing of the extracted point cloud data set to provide the substantially isolated point cloud data set of the structure includes performing data clustering.

7. The method according to claim 6, wherein the data clustering is region growing clustering.

8. The method according to claim 1, wherein the one or more predefined portions of the isolated point cloud data set in which the features are expected to be present are identified based on one or more predefined relative geometric positions associated with the structure within the isolated point cloud data set.

9. The method according to claim 1, wherein the predefined geometric models are preselected to fit shapes at or proximate the one or more predefined portions.

10. The method according to claim 9, wherein the predefined geometric models include one or more of cylinders, frusto-conical shapes and rectangular members.

11. The method according to claim 1, wherein the fitting of the predefined geometric models to one or more predefined portions of the isolated point cloud data set in which the features are expected to be present identifies and classifies associated isolated point cloud data as being representative of the fitted feature.

12. The method according to claim 11, wherein the identified and classified associated isolated point cloud data being representative of the fitted feature is at least temporary removed from the isolated point cloud data to facilitate efficient processing.

13. The method according to any one of the previous claims wherein the structure is a vertically projecting structure.

14. The method according to claim 1, wherein the structure is a tower standing on a ground surface.

15. The method according to claim 1, wherein the structure is a telecommunication tower with a pole supporting an antenna arrangement.

16. The method according to claim 15, wherein the one or more spatial parameters associated with the shape of the structure includes a spatial parameter representative of an estimated height of the tower.

17. The method according to claim 15, wherein the one or more predefined geometric models that represent features of the structure includes a pole model and an antenna model.

18. A system for generating a model of a structure from a point cloud data set of the structure, the system being configurable by software to:

Extract from the point cloud data set an extracted point cloud data set based on one or more spatial parameters associated with the shape of the structure;

Process the extracted point cloud data set to provide a substantially isolated point cloud data set of the structure;

Fit one or more predefined geometric models that represent features of the structure to one or more predefined portions of the isolated point cloud data set in which the features are expected to be present; and

Provide the model based on a combination of fitted ones of the one or more predefined geometric models.

19. A method for generating a model of a structure from a point cloud data set of the structure, the structure including a support feature and a peripheral feature carried by the support feature, the method including the steps of:

Fitting a first predefined geometric model that represents the support feature to a portion of the isolated point cloud data set in which the support feature is expected to be present and associating a first fitted portion of the isolated point cloud data set with the first predefined geometric model;

Further processing the substantially isolated point cloud data set of the structure to at least temporarily remove from processing the first fitted portion of the isolated point cloud data set to provide a reduced isolated point cloud data set; Fitting a second predefined geometric model that represents the peripheral feature to a further portion of the reduced isolated point cloud data set in which the peripheral feature is expected to be present and associating a second fitted portion of the isolated point cloud data set with the fitted second predefined geometric model;

Provide the model based on a combination of fitted ones of the first and second predefined models.

20. A method for processing a point cloud data set of a structure to provide a model of the structure, the structure including a support feature and a peripheral feature carried by the support feature, the method including the steps of:

Locating the structure within the point cloud data set and determining geometric bounds about the structure;

Extracting from the point cloud data set an extracted point cloud data set within the geometric bounds;

Processing the substantially isolated point cloud data set of the structure to provide a reduced isolated point cloud data set in which the first fitted portion of the isolated point cloud data set is removed;

Fitting a second predefined geometric model that represents the peripheral feature to a further portion of the reduced isolated point cloud data set in which the peripheral feature is expected to be present and associating a second fitted portion of the isolated point cloud data set with the fitted second predefined geometric model;

Providing the model based on a combination of fitted ones of the first and second predefined models.

21. A method for processing a point cloud data set of an object to provide a model of the object, the object including a hierarchy of features of varying geometric complexity, the method including the steps of:

Locating the object within the point cloud data set and determining geometric bounds about the object; Extracting from the point cloud data set an extracted point cloud data set within the geometric bounds;

Fitting a first predefined geometric model that represents at least a first of the hierarchy of features to a portion of the isolated point cloud data set in which the first of the hierarchy of features is expected to be present and associating a first fitted portion of the isolated point cloud data set with the first predefined geometric model;

Processing the substantially isolated point cloud data set of the object to provide a reduced isolated point cloud data set in which the first fitted portion of the isolated point cloud data set is removed;

Fitting a second predefined geometric model that represents at least a second of the hierarchy of features to a further portion of the reduced isolated point cloud data set in which the second of the hierarchy of features is expected to be present and associating a second fitted portion of the isolated point cloud data set with the fitted second predefined geometric model;

22. A method for sequentially processing a point cloud data set of an object to provide a computer aided drafting (CAD) model of the object using a set of predetermined models to represent at least some of a hierarchy of features associated with the object, the method including: sequentially fitting each of the predetermined models to the point cloud data and upon fitting of each of the predetermined models, removing data associated with fitted ones of the predetermined models to provide sequentially reduced point cloud data for processing and fitting of further ones of the predetermined models.

23. A method for sequentially processing a point cloud data set representing an object to provide a computer aided drafting (CAD) model of the object using a set of predetermined models to represent at least some of a hierarchy of features associated with the object, the method including: sequentially identifying point cloud data expected to be associated with each of the hierarchy of features, fitting each of the predetermined models to the identified point cloud data and upon fitting of each of the predetermined models, removing data associated with fitted ones of the predetermined models to provide sequentially reduced point cloud data for identification of further ones of the hierarchy of features and fitting of further ones of the predetermined models thereto.