Method for detecting vegetation on a railway track

Abstract

Disclosed is a method for detecting vegetation on a longitudinal segment of a railway track comprising at least some stock rails, at least one contact wire for supplying power to a railway vehicle and vegetation, the longitudinal segment being represented by a point cloud (N), the method comprising steps of determining wire points (Pf) corresponding to the contact wire and vegetation points (Pv) corresponding to the vegetation by means of a classification algorithm, of determining, for each vegetation point (Pv), a separation (e) corresponding to the shortest distance between the vegetation point (Pv) and one of the wire points (Pf), and of determining the number of vegetation points (Pv) associated with a separation (e) less than a first distance threshold (Sa), so as to estimate a maintenance requirement of the longitudinal segment of the railway track.

Description

Method for detecting vegetation on a railway track

[0001] The present invention relates to the field of railway track maintenance, in particular, the management of vegetation in the vicinity of a railway track.

[0002] A railway track generally comprises rails for the movement of a railway vehicle and a contact wire for supplying electricity to the railway vehicle. As is known, the contact wire is part of a catenary which is suspended above the rails by a plurality of poles.

[0003] To allow unhindered movement of the railway vehicle on the railway track, it is necessary to maintain the vicinity of the railway track. Generally, vegetation grows in the vicinity of the railway track and the latter is likely to obstruct the view of the railway vehicle driver, hit the railway vehicle during its movement, damage the contact wire and cause a break in the power supply, which is a critical event.

[0004] In order to eliminate these inconveniences, it is known to carry out maintenance of the railway track by an intervention team in order to eliminate the vegetation.

[0005] For these purposes, it is known to carry out visual monitoring of the railway track to determine the longitudinal sections of the railway requiring priority maintenance. Such monitoring depends on the operator and remains very subjective. It is also imprecise, complex to carry out and very expensive due to the human resources required. As a result, monitoring cannot be carried out at a high frequency, which is problematic.

[0006] The invention thus aims to eliminate at least some of these drawbacks, by proposing a new method for detecting vegetation on a railway track. PRESENTATION OF THE INVENTION

[0007] The invention relates to a method for detecting vegetation on a longitudinal portion of a railway track comprising at least running rails, at least one electrical power supply contact wire of a railway vehicle and vegetation, the longitudinal portion being represented by a point cloud, each point of the point cloud being associated with a volume unit, the method comprising steps consisting of:Determining, in the point cloud, the wire points corresponding to the contact wire and the vegetation points corresponding to the vegetation by a classification algorithm,Determining, for each vegetation point, a gap corresponding to the shortest distance between the vegetation point and one of the wire points,Determining the number of vegetation points associated with a gap which is less than a first distance threshold so as to estimate a maintenance requirement for said longitudinal portion of the railway track.

[0008] Such a method makes it possible to automate the detection of vegetation close to the contact wire, and thus eliminates the need for visual detection by operators. The method thus allows for high-precision vegetation detection with great objectivity. The method also makes it possible to estimate the maintenance requirement for each longitudinal section in a localized manner. This makes it possible to precisely target the longitudinal sections of the railway requiring maintenance, and thus facilitates the work of maintenance teams. The use of a point cloud advantageously allows for a three-dimensional representation of each longitudinal section, and thus allows for better automatic identification of vegetation than from a two-dimensional image.

[0009] According to one aspect, the method comprises a preliminary step of acquiring a point cloud of the longitudinal portion of the railway track by at least one three-dimensional acquisition sensor mounted on a railway vehicle traveling on the longitudinal portion. This allows the acquisition to be carried out in a practical, rapid, and automatic manner. In addition, this allows all the points of interest to be acquired, in particular the wire points and the vegetation points close to the contact wire. This also allows for easily repeatable acquisition, and thus good updating of the point cloud.

[0010] In one aspect, the three-dimensional acquisition sensor is a photogrammetric system, preferably allowing 3D reconstruction of the environment. This ensures good acquisition quality at low cost.

[0011] In one aspect, the method comprises a step of sub-sampling the point cloud so that each point is associated with a predetermined volume, preferably one cubic centimeter. This makes it possible to reduce the weight of the point cloud and thus increase the speed of its processing. This also makes it possible to easily approximate the vegetation density of the longitudinal portion.

[0012] According to one aspect, the classification step comprises sub-steps consisting of:Determining the wire points corresponding to the contact wire by an automatic classification algorithm, andDetermining the vegetation points corresponding to the vegetation by an automatic classification algorithm, the other points being points of the environment.

[0013] This advantageously increases the robustness of the classification of points, while maintaining a suitable execution time.

[0014] According to one aspect, the method comprises a step of correcting the classification of the contact wire points according to the height of said wire points. This makes it possible to discriminate contact wire points that are too low.

[0015] According to one aspect, the method comprises a step of correcting the classification of the vegetation points according to their alignment with wire points.

[0016] According to one aspect, the method comprises a step of correcting the classification of vegetation points if they are isolated from other vegetation points. Inconsistencies are thus eliminated.

[0017] This increases the robustness of the classification, and thus increases the robustness and accuracy of the maintenance requirement estimate. This correction step avoids mobilizing the maintenance team in the event of an error, and thus minimizes track maintenance costs.

[0018] In one aspect, vegetation points associated with a gap that is less than the first distance threshold are associated with a high risk. This increases the accuracy of the maintenance requirement estimate. This advantageously allows maintenance teams to target and prioritize the longitudinal sections with the most vegetation points associated with a high risk.

[0019] The invention also relates to a method for maintaining a railway track comprising a plurality of longitudinal portions, the method comprising:The steps of the method described above so as to determine for each longitudinal portion the number of vegetation points associated with a high risk, andA step of maintaining the longitudinal portions having the greatest number of vegetation points associated with a high risk.

[0020] Maintenance teams can therefore intervene only when necessary, prioritizing the longitudinal sections with the highest risk. PRESENTATION OF FIGURES

[0021] The invention will be better understood upon reading the following description, given by way of example, and referring to the following figures, given by way of non-limiting examples, in which identical references are given to similar objects.

[0022] This is a schematic representation of a side view of the railway track on which a railway vehicle is running.

[0023] This is a schematic representation of a front sectional view of the railway track.

[0024] This is a schematic representation of a top view of the railway track with several longitudinal sections.

[0025] This is a schematic representation of the transfer of the point cloud between the three-dimensional acquisition sensor and a computer.

[0026] This is a schematic representation of the vegetation detection process.

[0027] This is a schematic representation of a front sectional view of the point cloud acquired on the railway track.

[0028] This is a table showing the number of vegetation points in each longitudinal section according to their associated risk.

[0029] It should be noted that the figures set out the invention in detail to implement the invention, said figures can of course be used to better define the invention if necessary. DETAILED DESCRIPTION OF THE INVENTION

[0030] With reference to the, there is shown schematically a railway track 1 for the circulation of one or more railway vehicles 2. As described previously in the preamble, the railway track 1 comprises running rails 3 on which the railway vehicle 2 runs, and a contact wire 41 for supplying electricity to the railway vehicle 2 suspended above the running rails 3, configured to be in contact with a pantograph of the railway vehicle 2. In practice, as illustrated in the, the contact wire 41 belongs to a catenary 4. The catenary 4 comprises in a known manner a support wire 42 suspended above the rails by posts 43, the contact wire 41 being supported by the support wire 42.

[0031] In a manner similar to a highway, each longitudinal portion of a railway track 1 is identified from a determined distance from a reference point. In this example, with reference to the, the railway track 1 is divided into several consecutive longitudinal portions 10A, 10B, 10C so as to facilitate their identification by a maintenance team. In this example, the longitudinal portions 10A, 10B, 10C are centered on the running rails 3 and have a width of 15 meters for a length of 100 meters. Preferably, they have a height of 30 m above the contact wire 41. The predetermination of the dimensions of a longitudinal portion 10A, 10B, 10C makes it possible to know their volume V. As will be presented later, this makes it possible to determine the vegetation density 5 of each longitudinal portion 10A, 10B, 10C.

[0032] As indicated in the preamble, vegetation 5 grows in the vicinity of railway track 1. With reference to the, vegetation 5 can be in the form of trees, bushes, etc. Subsequently, vegetation 5 will be classified according to three risk levels: high risk Ra, moderate risk Rb or low risk Rc. It goes without saying that the number of levels could be different depending on the needs of the maintenance teams.

[0033] Subsequently, the risk level depends on the distance separating vegetation 5 from contact wire 41. The closer vegetation 5 is to contact wire 41, the greater the risk of cutting off the power supply to railway vehicle 2. The repair requires the intervention of a technical team, blocking traffic on railway track 1, which is a critical event.

[0034] For example, a tree extending over railway track 1 and possibly coming into contact with catenary 4 presents a high risk Ra, while a tree starting to grow towards railway track 1 presents a moderate risk Rb. Vegetation 5 growing towards the outside of railway track 1 presents a low risk Rc for catenary 4.

[0035] In order to schedule maintenance and avoid interruptions to traffic and power supply on the railway track 1, a vegetation detection system is proposed which comprises a three-dimensional acquisition sensor 6, mounted on the railway vehicle 2, making it possible to acquire a point cloud N of the railway track 1. It goes without saying that the railway vehicle 1 could comprise several three-dimensional acquisition sensors 6, for example three, so as to increase the robustness of the acquisition. Preferably, the three-dimensional acquisition sensor 6 is a photogrammetric system allowing 3D reconstruction of the environment which has high precision and reduced cost.

[0036] The acquired point cloud N advantageously allows a three-dimensional representation of the railway track 1. In addition, this allows point-by-point processing of the environment of the railway track 1, in order to detect and categorize the different zones of the railway track 1, in particular, the dangerous zones.

[0037] Preferably, the three-dimensional acquisition sensor 6 is associated with a geolocation module so that the points P of the point cloud N comprise geographic coordinates (x, y, z), which makes it possible to associate them with longitudinal portions 10A, 10B, 10C of the railway track 1.

[0038] According to one aspect, in addition to the geographic coordinates (x, y, z), each point P comprises an intensity of the return signal in response to the transmission signal of the three-dimensional acquisition sensor 6 and, optionally, a color, an echo level or other. According to one aspect, geometric attributes are calculated, in particular, linearity, flatness, dispersion, verticality, omnivariance and curvature. These attributes make it possible to locally describe and differentiate the points P. These attributes are determined from the coordinates (x, y, z) of each point P and the coordinates (x, y, z) of the nearest neighboring points. It goes without saying that the attributes could be different.

[0039] A method for detecting and classifying vegetation 5 automatically using a CAL computer connected to the three-dimensional acquisition sensor 6 as illustrated in the will now be presented. The CAL computer can be on-board or remote.

[0040] With reference to the, the method comprises a step E1 of acquiring a point cloud N of the longitudinal portion 10A, 10B, 10C of the railway track 1. Preferably, the acquisition E1 of the points P of the point cloud N is done by using one or more three-dimensional acquisition sensors 6 on board a railway vehicle 2, as described previously. This makes it possible to acquire points P without interrupting traffic on the railway track 1, while ensuring high precision of the acquired points for their processing during the following steps. In addition, unlike an acquisition carried out by an airborne three-dimensional acquisition sensor, this makes it possible to ensure the acquisition of the points P representative of the contact wire 41 and the vegetation 5.

[0041] The railway vehicle 1 carries out the acquisition step E1 periodically, so as to update the point cloud N and allow an update of the vegetation detection 5. Each point P corresponds to a three-dimensional volume which is a function of the precision of the three-dimensional acquisition sensor 6.

[0042] A method for processing the points P of the point cloud N will now be presented in order to estimate the maintenance requirement.

[0043] In order to speed up the processing, the method may comprise a step E2 consisting of sub-sampling the point cloud N so as to limit the number of points P to be processed. Preferably, the sub-sampling makes it possible to limit the number of points in the point cloud N so as to keep one point for a certain unit of volume. In this example, one point P is kept per cm3. The sub-sampling step E2 is optional and depends on the available computing power of the CAL calculator.

[0044] The method comprises a step consisting of classifying E3 the points P of the point cloud N so as to determine the vegetation points Pv.

[0045] A classification algorithm assigns a class to each point P of the point cloud N. Each point P is thus classified as being a wire point Pf, a vegetation point Pv, or an environment point Pe. In this example, the wire points Pf corresponding to the contact wire 41 and the vegetation points Pv are determined by an automatic classification algorithm from the attributes of the points Pi. In this example, the environment points Pe correspond to everything that is not the contact wire 41 or the vegetation 5, such as for example the traffic rails 3, the support wire 42, the electric poles 43 or the fences near the railway 1.

[0046] The classification algorithm is preferably the “Random Forest” algorithm known to those skilled in the art. It is previously trained on a data set comprising wire points Pf and vegetation points Pv.

[0047] Advantageously, the number of vegetation points Pv in each longitudinal portion 10A, 10B, 10C is thus determined, which makes it possible to determine a vegetation density for each longitudinal portion 10A, 10B, 10C from the known volume of each longitudinal portion 10A, 10B, 10C. The vegetation density allows the maintenance teams of railway track 1 to better plan and better target their interventions.

[0048] Optionally, the method includes a step E4 consisting of correcting the classification of the points P based on business rules. For example, the height of the wire points Pf is checked to verify that they have a height greater than a predetermined threshold value, for example, 2.5 meters. Otherwise, the wire points Pf are corrected as environment points Pe.

[0049] According to one aspect, a RANSAC “RANdom SAmple Consensus” algorithm, known to those skilled in the art for forming coherent sets of points, is applied to the wire points Pf to determine the wire points Pf corresponding to small objects and not being able to correspond to the contact wire 21. They are then classified as environment points Pe.

[0050] According to one aspect, the sets of points detected by the RANSAC algorithm are used to determine the sets of vegetation points Pv strictly in alignment with the wire points Pf in order to classify them as wire points Pf. Thus, a continuity of wire points Pf is determined to represent a contact wire 41.

[0051] Furthermore, sets of isolated or "floating" vegetation points Pv, i.e. those which are not connected to other vegetation points Pv, are classified as environment points Pe.

[0052] The environment points Pe can be removed from the point cloud N in order to reduce the weight of the point cloud N or kept in memory in order to preserve a complete image of the railway track 1 at the time of acquisition of the point cloud N, in particular for archiving or presentation purposes.

[0053] The method comprises a step consisting of calculating E5, for each vegetation point Pv, a gap Pv(e) corresponding to the shortest distance between said vegetation point Pv and one of the wire points Pf.

[0054] In this example, the calculation is carried out using a “kd” tree type algorithm, known to those skilled in the art, from the wire points Pf. This algorithm makes it possible, for each vegetation point Pv, to determine the closest wire point Pf. The distance Pv(e) is then calculated for each vegetation point Pv from the coordinates (x, y, z) of said vegetation point Pv and the coordinates (x, y, z) of the closest wire point Pf, by a Euclidean distance calculation. Thus, for each vegetation point Pv, there is a distance Pv(e).

[0055] As illustrated in, the method comprises a step consisting of determining E6 the number of vegetation points Pv associated with a deviation Pv(e) which is lower:than a first distance threshold Sa corresponding to a high risk Ra,than a second distance threshold Sb corresponding to a moderate risk Rb,than a third distance threshold Sc corresponding to a low risk Rc.

[0056] In this example, the first distance threshold Sa is equal to 1 meter, the second distance threshold Sb is equal to 2 meters and the third distance threshold Sc is equal to 3 meters. It goes without saying that the number of distance thresholds / risk levels could be adapted.

[0057] Thus, each vegetation point Pv is associated with a risk level Ra, Rb, Rc. This advantageously makes it possible to determine for each longitudinal portion 10A, 10B, 10C the number of vegetation points Pv as well as the associated risk level as illustrated in the table. In this example, the longitudinal portion 10B is the first priority because it has the largest number of vegetation points Pv associated with a high risk Ra. The longitudinal portion 10C is the second priority because it has the largest number of vegetation points Pv associated with a moderate risk 10B.

[0058] In practice, it is sufficient for a railway vehicle 2 to travel on the railway track 1 to acquire a point cloud N during its travel which can be processed automatically in order to determine, for each longitudinal portion 10A, 10B, 10C, the number of vegetation points Pv as well as the associated risk level Ra, Rb, Rc.

[0059] Thus, a maintenance team automatically has qualitative, precise and objective information to organize the maintenance of the highest priority longitudinal sections 10A, 10B, 10C. The detection process facilitates the work of maintenance teams by precisely locating the longitudinal sections 10A, 10B, 10C requiring intervention. Detection can be renewed periodically in order to have updated and relevant data.

Claims

Method for detecting vegetation on a longitudinal portion (10A, 10B, 10C) of a railway track (1) comprising at least running rails (3), at least one contact wire (41) for supplying electricity to a railway vehicle (2) and vegetation (5), the longitudinal portion (10A, 10B, 10C) being represented by a point cloud (N), each point (P) of the point cloud (N) being associated with a unit of volume, the method comprising steps consisting of:Determining (E3), in the point cloud (N), the wire points (Pf) corresponding to the contact wire (41) and the vegetation points (Pv) corresponding to the vegetation (5) by a classification algorithm,Determining (E5), for each vegetation point (Pv), a gap (e) corresponding to the shortest distance between the vegetation point (Pv) and one of the wire points (Pf),Determine (E6) the number of vegetation points (Pv) associated with a gap (e) which is less than a first distance threshold (Sa) so as to estimate a maintenance requirement for said longitudinal portion (10A, 10B, 10C) of the railway track (1)., Method according to claim 1, comprising a preliminary step consisting of: Acquiring (E1) a point cloud (N) of the longitudinal portion (10A, 10B, 10C) of the railway track (1) by at least one three-dimensional acquisition sensor (6) mounted on a railway vehicle (2) traveling on the longitudinal portion (10A, 10B, 10C). Method according to claim 2, in which the three-dimensional acquisition sensor (6) is a photogrammetric system. Method according to one of claims 1 to 3, comprising a step consisting of: Sub-sampling (E2) the point cloud (N) so that each point (P) is associated with a predetermined volume, preferably one cubic centimeter. Method according to one of claims 1 to 4, in which the classification step (E3) comprises sub-steps consisting of:Determining the wire points (Pf) corresponding to the contact wire (41) by an automatic classification algorithm, andDetermining the vegetation points (Pv) corresponding to the vegetation (5) by an automatic classification algorithm, the other points (P) being points of the environment (Pe). Method according to one of claims 1 to 5, comprising a step of correcting the classification (E4) of the contact wire points (Pf) as a function of the height of said wire points (Pf). Method according to one of claims 1 to 6, comprising a step of correcting the classification (E4) of the vegetation points (Pv) as a function of their alignment with wire points (Pf). Method according to one of claims 1 to 6, comprising a step of correcting the classification (E4) of the vegetation points (Pv) if they are isolated from other vegetation points (Pv). Method according to one of claims 1 to 8, in which the vegetation points (Pv) associated with a gap (e) which is less than the first distance threshold (Sa) are associated with a high risk (Ra). Method for maintaining a railway track (1) comprising a plurality of longitudinal portions (10A, 10B, 10C), the method comprising:The steps of the method according to claim 9 so as to determine for each longitudinal portion (10A, 10B, 10C) the number of vegetation points (Pv) associated with a high risk (Ra), andA step of maintaining the longitudinal portions (10A, 10B, 10C) having the greatest number of vegetation points (Pv) associated with a high risk (Ra).

Images