WO2022152973A1

WO2022152973A1 - A method for identifying a road type

Info

Publication number: WO2022152973A1
Application number: PCT/FI2022/050023
Authority: WO
Inventors: Vesa Leppänen
Original assignee: Oy Arbonaut Ltd.
Priority date: 2021-01-13
Filing date: 2022-01-12
Publication date: 2022-07-21
Also published as: EP4278214A1

Abstract

A system and method of identifying a road type. In the method the source data is collected by acquiring a point cloud using airborne remote sensing sensor. The point cloud is then processed so that the road type can be determined. The road type can be measured from a profile that is generated from the acquired point cloud and received additional parameters. Instead, or in addition to the road type, also road surface and road ditch can be measured.

Description

A METHOD FOR IDENTIFYING A ROAD TYPE

DESCRIPTION OF BACKGROUND

The following disclosure relates to identifying a road type based on point cloud data . Particularly, the disclosure relates to a system and method of extracting a profile of a road from a point cloud and identifying the type of the road from the profile .

Identifying the construction type of a road is important for automatic analysis of roads . There are many reasons why organi zations and individuals are interested in the road construction type including but not limited to :

- Making decisions on road maintenance need;

- Designing road improvements ;

- Quantifying amounts of earth movement need on road improvements ;

- Si zing road components , like culverts , fences or siderails .

Some examples of operations where road type identification is interesting for the organi zations or individuals :

- Automatic analysis of road improvement need . Different road construction types are applied on different road construction conditions . The dimensions , excavation instructions and material quantities are different for different road profile types . This leads to differing construction, maintenance and improvement costs .

- Timber transport and other use of lower level road network is dependent on road usability . The road construction prof ile has an impact on the usabil ity of the road during different seasons . For example , timber stumpage price tends to alter depending on the traf fic conditions to the harvest site . Road profile and the corresponding minimum dimensions of the profile allow j udgment of the traffic conditions on a given road .

Buying property; As on timber purchase , traffic conditions have an effect on the purchase price of a given property .

Making decisions on subsidies . Roads , providing benefits to large part of the societies , tend to be common target of public-private partnerships . Similarly, public subsidies are commonly awarded for road maintenance and construction . The decisions to award subsidy for a road maintenance or improvement as well as decision to approve made improvements for subsidy payments may require information about road profile and dimensions either before improvement or after it .

Remote sensing is an art known to mankind . Remote sensing may be performed using for example satellite or airborne sensors , operated from manned or unmanned vessels . Remote sensing has been done from land and water vehicles as well as from airborne vessels or spacecraft . Sensors most commonly used include spectral sensors ( cameras , spectrometers etc . ) , LiDAR sensors and radar sensors , but other kinds are known to be used as well .

Remote Sensing has capability to produce information about obj ects , like roads and vegetation as well as their geometry . This information may be geographically two-dimensional or three-dimensional , but can also include more dimensions , like time . Some examples of information related to road management , achieved from remote sensing, include the road center line location and road width, mentioned above .

Three-dimensional point clouds have been used quite extensively in sensing . Point clouds can be produced using multiple techniques , LiDAR, photogrammetry and radargrammetry being j ust a few . A brief presentation of some of the techniques are presented here .

Photogrammetry is the science of making measurements from photographs . Stereophotogrammetry is a methodology of Photogrammetry where group of two or more images taken of the same target are analyzed . The images are taken from different viewpoints , presenting the obj ects at different distance from the observing sensors at different locations in the imaging sensor . Corresponding features are identified in different images and their relative location on the image are interpreted to extract the 3D location of the obj ects . The sensor locations may be given to the algorithm or, alternatively, deduced from the analysis .

LiDAR, known also as laser scanning, has been used for forest inventories approximately since 1990 ' s , but somewhat longer time in topographic analysis . LiDAR is an active instrument that uses laser ranging, combined with devices measuring position and attitude of the sensor, to produce 3D location measurements of obj ects . The sensor emits a laser beam to a known direction from a known position and records the distance to surfaces where the beam is reflected back . Additionally, LiDAR may have capability to record the intensity of the returning signal , indicating the reflectivity and si ze of the reflecting surfaces . The laser beam is proj ected to the obj ect through a mirror or prism system or other kind of optical setup ( the "LiDAR Optic" ) that causes the laser beam to scan the target area, recording the precise direction where the beam was sent each time to allow construction of the 3D measurements .

LiDAR has been further developed to use an array of laser beams instead of a single beam . The array may be stationary or scan the targeted area . However, the system yields a set of three-dimensional coordinates and potentially some information of the reflectivity . Information from this kind of a sensor is substantially similar to the information received from a traditional single-beam LiDAR and embodiments presented in this publication are applicable as such .

LiDAR has been improved also by adding lasers of different light bandwidth . These sensors are capable of measuring the intensity of the returning pulses at different bandwidths , and can yield information about the target reflectivity on different bandwidths . Despite this additional spectral information, the data is similar to the data from traditional single beam LiDAR, and can be processed as such in the presented process .

Traditionally, LiDAR has been used to produce attributes to areas of land . For example , LiDAR-derived attributes have been assigned to timber stands , making management or inventory units .

Radargrammetry is a technology of extracting geometric obj ect information from radar images . The output of the radargrammetric analysis may be for example a geometric three-dimensional point cloud . Like stereophotogrammetry and LiDAR, also radargrammetry can be used from airborne or satellite , ground and water vessel platforms .

Thus , there is a need for improving methods for classifying road construction type .

SUMMARY

This disclosure discloses an approach, method and a process to identify Road Construction Type automatically from a point cloud .

The method differs from the known methods known to mankind before by presenting an automated approach to identify the Road Construction Type where other methods of identification of the Road Construction Type may be inaccurate , impractical or costly to acquire . While many measurements may be obtained from point clouds manually, it is time consuming and costly to perform such measurements in a large scale . Automatic analysis of point clouds , obtained by sensors like LiDAR, Radar or photogrammetry would be a useful way to obtain information from existing roads . However, most analysis results are useful only in the context of the Road Construction Type . For example , the ditch depth measurement or road surface width measurement are useful if they are connected to the construction type . Defining the construction type and the profile of a road is a challenge in automatic analysis .

In an aspect a method of identifying Road Construction Type automatically from a point cloud is disclosed . The method comprises receiving center line of the road of interest ; receiving point cloud; producing geometric Analysis Polygons related to the center line ; further dividing Analysis Polygons to Profiling Polygons ; generating a profile comprising at least two profiling polygons and their locations ; and identifying at least one of the following based on the generated prof ile : the road type , the road surface and roadside ditch .

It i s beneficial to use a method as disclosed in this document for identifying Road Construction Type automatically from a point cloud . The method provides additional information for further analyzing the road which may be taken into account in decisions . Additionally, the method provides means for improving the road management by identifying the points that may need maintenance so that they can be improved to maintain the transport capacity of the roads .

In an implementation the point cloud is data received from remote sensing . It is beneficial to use remote sens ing as it provides an easy way of acquiring information covering large areas of terrain .

In an implementation at least a portion of the point cloud is derived from data acquired by LiDAR sensor . In another implementation at least a portion of the point cloud is derived from data acquired by photogrammetry . It is beneficial to use known approaches for remote sensing as they provide reliable information . Furthermore , in some implementations it may be useful to combine these two and possibly with additional sensing mechanism .

In an implementation the method further comprises receiving road class , producing the said Analysis Polygons or Profiling Polygons according to the received road class ; still summari zing point cloud information within the produced Analysis Polygons or Profiling Polygons ; and identifying Road Construction Type automatically from a point cloud based on summari zed information . Identifying the profile type may depend on the road class ; for example , higher profile may be required from a certain road class A to qual ify a filled road than from another road class B . It is benef icial to use the road class in conj unction of the Road Construction Type in the decision making .

In an implementation the summari zing of the vegetation information is done using an optimi zation algorithm to define the Road Construction Type . Training data may be used to introduce the Road Construction Type samples . I f each sample is connected to a Road Construction Type , some optimi zation algorithm may be able to find the Profile that corresponds to each Road Construction type and classify any set of samples to those given types .

In an implementation the method further comprises using artificial intelligence to perform the classification based on training data . The Profile information and derivatives may be provided as independent variables to Al .

In an implementation the method further comprises Identifying the road surface by fitting a curve to the profile and detecting a deviation from the curve . In another implementation the method further comprises identifying the roadside ditch by first detecting the road surface and then detecting the profiling polygon with the lowest value of the elevation statistic outside the road surface .

In an aspect a system compris ing at least one processor and data communication connection is disclosed . The system is configured to perform a method disclosed above .

In an aspect a computer program comprising computer program code is disclosed . The computer program is configured to cause a computing device to perform a method as disclosed above when the computer program is executed in a computing device .

The above disclosed methods , system and computer program improve identifying Road Construction Type automatically from a point cloud . The classified information can be used to improve road maintenance , management and improvements .

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings , which are included to provide a further understanding of the system and method to identify road construction type on point cloud data and constitute a part of this specification, illustrate embodiments and together with the description help to explain the principles of the system and method of identifying road construction type on point cloud data . In the drawings :

Fig . 1 presents a flow chart of an example embodiment of the process of identifying road construction type on point cloud data ;

Fig . 2 presents a block diagram of an example of an arrangement for identifying a road type ;

Fig . 3 presents a cross-sectional profile viewed in a point cloud; Fig . 4 presents a cross-sectional profile viewed in a point cloud; and

Fig . 5 presents a cross-sectional profile viewed in a point cloud .

DETAILED DESCRIPTION

Reference will now be made in detail to the embodiments , examples of which are illustrated in the accompanying drawings .

In the following description system and method of identifying road construction type on point cloud data is disclosed .

The Figure 1 shows an example of a case where road construction type is defined automatically from a point cloud, some useful measurements are performed and connected to the context of the construction type for making useful decisions . In the example process , point cloud data, step 101 , is obtained using sensing with some sensor A . The received point cloud may be used as it is or it may be pre-processed and/or classified, step 103 , before using it in the process at step 108 . Sensor a may be operating from an airborne vehicle over the target road . Sometimes , point clouds are classified and consist of points that are classified to different classes . This may be the case , for example , on LiDAR data, where it is common to indicate the type of surface that the point represents by a class attribute assigned to the point . Assigning this class is called Point classification . For example , some point cloud on road area may contain points representing ground surface and some vegetation in the vicinity of the road . It may be practical to classify the points and feed only the points representing the surface of the earth to the road analysis process . For example , the sensor may be a LiDAR device carried by a drone . Other suitable sensors and air carriers may be used . The data from the sensor is turned into a point cloud, each point representing a return from the ground surface . Inside the point cloud area, there is the target road R . The center line of the road is received, step 102 , approximately represented by a vector polyline ("Centerline Vector" ) or a feature of similar effect . The center line of the road is received from a road database . The road database may be provided by government , other public organi zation or a private company, or identified manually or automatically from some data . Typically, the road databases comprise road information in high accuracy form . It is possible, however, that the center line of the road is not accurate . It is possible , that the road center line location does not overlay within the area where the road is identi fied in the point cloud data . In this kind of a case , it may be practical to improve the road center line location before entering it as an input to the process of identifying road construction type on point cloud data . Methods are available to detect road center line location from remote sensing data, Including some spectral analysis and geometric analysis of point cloud data . Additionally, methods exist to improve road center line location automatically on remote sensing data, some of them being analysis of cross-sectional profile data and other being analysis of spectral information or ground surface roughness or any combination of methods to detect or improve road location information . Without going into details of these identification, detection or accuracy improvement methods , it is possible to apply any such method to improve the road centerline location before using the method to identify road construction type on point cloud data .

The Centerline Vector' s task is to input the information about the location of the road of interest . The Centerline Vector is then divided into "Analysis Units" by dividing it to pieces , step 105 . Each Analysis Unit is approximately of certain length L and has some direction D and a unique ID and the direction allowing to define left and right . A practical length of an Analysis Unit may be 2 -20 meters . Each Analysis Unit is turned into an "Analysis Polygon" ( 16 ) by adding a buffer to both sides , step 106 . A practical buffer si ze is 5-20 meters , however, also other buffer sizes may be used . In an implementation the Analysis Polygons are formed from the Road Center Line using any other method, essentially, reaching a substantially similar outcome . Each Analysis Polygon is given a unique ID of the originating Analysis Unit and further sliced to "Profiling Polygons" by dividing it in direction of the Analysis Unit into polygons , step 107 , wherein each Profiling Polygon having a width of W . The ID of the Analysis Unit and Analysis Polygon where it is originated from is attributed to each of the Profi ling Polygons , similarly, the Profiling Polygons are indexed according to their position related to the center line , for example , from left to right . A practical width W of a Profiling Polygon is 0 . 1 -2 meters . The optimal Profiling Polygon width depends on the density of the point cloud and the amount of detail desired from the analysis . For each Profiling Polygon, the ground elevation is defined by taking some statistic Z from the point cloud, step 108 . A practical statistic may be the minimum elevation, mean elevation, median elevation or value of some percentile point of the elevations of the point cloud points landing inside the Profiling Polygon, but may also be some statistic of the color value or spectral intensity value . "The profile" of the crosssection of the road can be presented by plotting the statistic Z values of all Profiling Polygons in one Analysis Polygon in the order of the Profi ling Polygon indexes . The Profiling Polygons may be divided to di fferent zones based on their location related to the Analysis Unit they are made from, step 110 . For example , the following zones may be identified : Center Area ; Profiling Polygons in the proximity of the center line . The center area may be further defined as a set of Profiling Polygons withing a buffer of B from the Analysi s Unit , with the Z value differing less than threshold D from the Z value of the Profiling Polygons closest to Analysis Unit . There are many ways to detect the Center Area ; including but not limited to gradient analysis of Z value , thresholding of Z value etc .

Left bank; some number of the most extreme Profiling Polygons on the left side of the center line .

Right bank, some number of the most extreme Profiling Polygons on the right side of the center line .

The location of the center area, left bank and right bank may be better defined by first identifying the road surface . Road surface may be detected by any means practical to this purpose , one example being slicing each Analysis Unit to narrow, co-directional polygons in the direction of the road center line . The elevation of each slice may be estimated by taking some statistic of the ground returns , for example , min, mean or median value of the z coordinate . When ordered across the road centerline direction, these surface elevation values present a road profile . For example , starting from the center line , it is possible to detect the point both left and right of the center line where the profile curves down or up from the road surface . In a further example , difference from the mean value of all slices from the center line to the analyzed slice can be compared and a threshold set to detect when the surface of the earth starts to curve down, indicating the edge of the structural road surface . More elegant detection methods may be used, including, but not limited to , derivate analysis , second derivate analysis , fitting a line to the values ( indicating road surface crosssection) and detecting a deviation from thi s l ine . The fitted line may be a direct line or some formula presenting theoretical or design profile of a road or even the road under analysis , if known . The important thing is to somehow detect the road surface from the slice elevations . When the road surface has been detected, the left and right bank detection may be done by selecting a distance from the road surface edge where the banks starts and another distance where it ends . Furthermore , it is possible to also detect the road ditch bottom, when existent , by finding the s lice with the lowest elevation outside the road surface on each side . I f the lowest elevation on one side of the road is the outermost analysis slice, it can be deduced that a ditch does not exi st in that s ide of the road, since any poss ible water would probably be shed out directly downhill , not following a ditch . I f a ditch is detected, it is possible to use the distance between the road surface and the ditch to define where the undisturbed bank might be ; for example , the distance between the road surface and the ditch may be multiplied with some parameter value to indicate where the undisturbed road bank might be located . Having undisturbed bank start from a distance from the road surface that is 2 or 3 times the distance between the road surface and the ditch bottom might be a good default value . In the analysis of road profiles , it is practical to have the road profile designs used in the target area as guides . The road design instructions often define the roadside grades and road ditch side grades , helping to find a suitable parametri zation . Furthermore , now that the distance between the road ditch and the road surface is known, if the elevation difference between the ditch bottom and the road surface is taken by using some statistic of the point cloud elevations in the slices that landed in the road surface and the sl ice that was determined to be the ditch bottom (min, mean, median, some percentile etc . being potentially suitable statistics ) , the grade of the roadside ditch can be measured. Optionally, this grade may be useful in determining the grade of the outer side grade of the ditch; after all, the soil properties may decide the suitable grading. Grading, if substantial, is normally shaped to a straight, sloped surface. A point may be analyzed where the grading of the outer side of the ditch ends using similar approach as described earlier for detection of the edge of the road surface: for example, fitting a line through the profile of the earth surface on the grade and finding the point where the surface deviates from the line, essentially, marking the outer edge of the graded surface.

The goal of the definition of the left and right bank is to position them to the area in the surrounding surface of the Earth, that is not manipulated by the road construction. If the outer edge of the graded area is detected as defined above, it is possible to define the bank as the area covered by the Profiling Polygons, outside the graded area. The width of the bank polygons can be set to some reasonable value, or just a given number of Profiling Polygons out from the edge of the grading can be used.

It may be useful, instead of analyzing the gradings, just to define the banks as a set of Profiling Polygons starting from a fixed distance from the roadside and extending another fixed distance out. In practical testing, this has proved to be effective and computationally economical. This kind of a fixed width might be, for example, an area extending from 4-10 m outside the edge of the road surface. It is also possible to use the road surface width as a guide and mark the bank area start from a distance from the road edge, where the distance is proportional to the road surface width. A suitable proportion might be, for example, 0,5- 1.5 x the road width.

Furthermore, it is possible to form the Profiling Polygons in such a way that they are not rectangles . They may be curved slices , following the road center line curvature , any road groove direction etc . I f such Profiling Polygons are used, it may be practical to define the bank areas also being an area formed by combining a set of these Profiling Polygons .

A logic may be appl ied to the prof ile to define the profile type ( construction type ) , for example the following way ( Table 1 ) :

Instead of having one left bank, one Center

Area and one right bank, it i s possible to perform the logic to define the road type , using several areas ; two or more left bank areas , two or more center areas and two or more right banks . I f several areas are defined, the logic may be similar as shown above , but having more operators ; for example , if the banks are defined with two Bank areas , the Right Bank Z statistic value on the closer Right Bank area may be higher than the road surface , but the outer may be lower, indicating a high bank but a generally decreasing landscape .

Additionally, if the only interest is on one side of the road, one of the banks may be omitted altogether .

The construction type may be attributed to each Analysis Unit , Analysis Polygon, Profiling Polygon, center area, left or right bank using the ID values that connect the geometries to their parents . Depending on the application, the usability of the information may differ based on the geometry it is connected . For example , in mapping applications , it may be practical to have the road construction type attributed to the Analysis Units that are presented with the center line . This center line may be colored with different color on the map depending on the construction type , making a useful map of road construction types .

In this example , each Analysis Unit of the target road is given a construction type according to the logic presented in the Table 1 . Logical operator > indicates that the value of the Z statistic is significantly higher than the value of the ZC; "significantly" meaning that the difference is bigger than a threshold TZ . Logical operator < indicates that the value of the Z statistic is significantly lower than the value of the ZC ; "Significantly" meaning that the difference is bigger than a threshold TZ . I f the Z statistics on both banks of the Center Area are within the threshold ZC from the Z , the road may be classified to "Flat Road" , step 109 .

The Road Construction Type may be further utili zed to define analysis types and parameters for continued analysis pertaining to road . For example , one interesting parameter about road condition is the depth of ditches . However, the ditch depth is defined depending on the Road Construction Type ; if the road is side-cut ( left-cut or right-cut construction type ) , only the depth of the ditch on the uphill side may be of interest ( the downhill side shedding water downhill without ditches ) , while on cut and filled roads both ditches are of interest . Furthermore , the effect of a ditch on a road of cut construction type is different from a road on a filled construction type . Thus , the depth measurement , while it may be performed in a similar manner, may be j udged in a different context . For example , similar ditch depth may indicate a good road structure on one construction type while being inadequate on another construction type .

The step as described above may include receiving the road class , step 104 . The road class may be used in several different way . For example , the road class may be received already in the beginning so that it is used when determining Analysis Units or it may be used only when doing the final classification, or in any step between . One practical use of the road class may be in defining the location of the banks ; different road classes having different construction dimensions and, thus , different bank locations .

Figure 2 discloses an example of an arrangement for performing a method as described above . The arrangement comprises a flying vehicle , 200 , and a processing system 201 . The flying vehicle comprises a measurement sensor 206 , which may be a LiDAR-device or similar . The measurement sensor 206 is configured to acquire a point cloud or other visual data that can be used in determining the road 202 . The point cloud is not limited to the road only but covers also at least the vicinity of the road . The flying vehicle may be a drone , helicopter, airplane or a satellite . The flying vehicle comprises all equipment necessary for flying . The collected measurement data is then transferred into the processing system 201 , such as a server, cloud service, work station or similar computing resource . The transmission may be performed by using telecommunication connection, such as a wireless local area network or mobile communication network . Instead of a network connection a transferrable media, such as a memory card or similar, may be used .

The processing system 201 comprises at least one memory 203 that is configured to store computer programs and relevant data, which includes data acquired by the flying vehicle 200 . The processing system 201 further comprises at least one processor 204 configured to execute computer programs and perform tasks caused by the instructions of the computer program . Furthermore , the processing system 201 comprises means for receiving information 205 , for example a memory card reader or a network interface .

The arrangement of figure 2 is then used for performing a method as discussed above with regard the example of Figure 1 .

Fig 3 . Road cross-sectional profile viewed in a point cloud . The direction of the road is heading away from the viewer . Road surface zone 301 , left bank zone 302 and right bank zone 303 are presented . Difference of elevation between road surface elevation 304 and bank elevation 305 is less than some threshold T ; so the road is classified to present a flat construction type .

Fig 4 . Road cross-sectional profile viewed in a point cloud . The direction of the road is heading away from the viewer . Road surface zone 401 , left bank zone 402 and right bank zone 403 are presented . Difference of elevation between road surface elevation 404 and bank elevation 405 is more than some threshold T , while left bank is higher than the road, right bank being lower than the road; so the road is classified to present a ( left ) side cut construction type .

Fig 5 . Road cross-sectional profile viewed in a point cloud . The direction of the road is heading away from the viewer . Road surface zone 501 , left bank zone 502 and right bank zone 503 are presented . Difference of elevation between road surface elevation 504 , left bank elevation 506 and right bank elevation 505 are more than some threshold T , whi le left and right banks are higher than the road; so the road is classified to present a cut construction type .

The above-described methods may be implemented as computer software which is executed in a computing device that can be connected to the Internet . When the software is executed in a computing device it is configured to perform a method described above . The software is embodied on a computer readable medium, so that it can be provided to the computing device .

As stated above , the components of the exemplary embodiments can include a computer readable medium or memories for holding instructions programmed according to the teachings of the present embodiments and for holding data structures , tables , records , and/or other data described herein . The computer readable medium can include any suitable medium that participates in providing instructions to a processor for execution . Common forms of computer-readable media can include , for example , a floppy disk, a flexible disk, hard disk, magnetic tape , any other suitable magnetic medium, a CD- ROM, CD±R, CD1RW, DVD, DVD-RAM, DVD1RW, DVD1R, HD DVD, HD DVD-R, HD DVD-RW, HD DVD-RAM, Blu-ray Disc, any other suitable optical medium, a RAM, a PROM, an EPROM, a FLASH-EPROM, any other suitable memory chip or cartridge , a carrier wave or any other suitable medium from which a computer can read .

It is obvious to a person skil led in the art that with the advancement of technology, the basic idea of the system and method of extracting a profile of a road from a point cloud and identifying the type of the road from the profile may be implemented in various ways . The system and method of extracting a profi le of a road from a point cloud and identifying the type of the road from the profile and its embodiments are thus not limited to the examples described above ; instead they may vary within the scope of the claims .

Claims

1. A method for identifying a road type comprising:

Receiving a point cloud from a measurement sensor;

Receiving a road center line;

Producing analysis units based on the received road center line;

Producing profiling polygons based on the produced analysis units;

Generating a profile comprising at least two profiling polygons and their locations; and

Identifying at least one of the following based on the generated profile: the road type, the road surface and roadside ditch.

2. The method according to claim 1, wherein the method further comprises:

Producing analysis polygons based on the produced analysis units and producing said profiling polygons based on produced analysis polygons.

3. The method according to claim 1 or 2, wherein the method further comprises:

Preprocessing the received point cloud.

4. The method according to claim 1 - 3, wherein the method further comprises:

Defining at least one zone within one or more analysis polygons .

5. The method according to claim 1 - 4, wherein the method further comprises:

Receiving a road class for the road being identified.

6. The method according to claim 1 - 5, wherein the method further comprises instructing a flying vehicle comprising a measurement sensor to acquire a point cloud . A method according to any of preceding claims 1 -

6 , wherein the method further comprises Identifying the road surface by fitting a curve to the profile and detecting a deviation from the curve . A method according to any of preceding claims 1 -

7 , wherein the method further comprises identifying the roadside ditch by first detecting the road surface and then detecting the profiling polygon with the lowest value of the elevation statistic outside the road surface . A computer program comprising computer executable code , which is , when executed on a computing device , configured to perform a method according to any of claims 1 - 8 . An apparatus comprising a processor and a memory, wherein the apparatus is configured to execute a computer program according to claim 9 .