Method and device for locating a point of interest on a plot of land from an image

Abstract

The invention relates to a method and a device for locating a point of interest on a plot of land from an image, by selecting a mesh (13-18) of a digital model of the plot of land comprising meshes tiling this plot of land, each mesh being associated with an altitude representative of the portion of the plot of land that it tiles, characterised in that it comprises the following steps: - obtaining an image of the plot of land taken by an optical sensor; - identifying and locating the point of interest in the image; - obtaining the position of the optical sensor at the time of capturing the image in a terrestrial reference frame — for example, altitude, longitude, latitude; - determining the orientation of a sighting axis (10) pointing towards the point of interest; - projecting the terrestrial position of the optical sensor onto the digital model of the plot of land in a predetermined direction of projection; - projecting the sighting axis (10) onto the digital model of the plot of land in the same predetermined direction of projection; - implementing the Bresenham algorithm to determine a mesh path (13-18) in the digital model of the plot of land, including the projection of the terrestrial position of the optical sensor and following the projection of the sighting axis; and - searching for the mesh of the path having an associated altitude which corresponds to the distance between the mesh and the projection of the mesh on the sighting axis, according to the predetermined direction of projection.

Description

Method and device for locating a point of interest on a terrain from an image

[0001] The present invention relates to the technical field of geolocation of points of interest on the Earth's surface, and in particular the field of combating vegetation fires.

[0002] Programs and algorithms have already been developed with the aim of enabling a drone equipped with a multimodal camera, i.e. operating in the visible and infrared spectrums, to carry out a mission automatically or remotely to geolocate points of interest. The mission consists of making it follow a trajectory above a plot of land based on several GPS points. During the flight over the plot, the drone is able, thanks to an onboard computer, to detect hot spots resulting from vegetation fires.

[0003] A digital terrain model, or DEM, is a digital representation of the relief of a geographic area. It is a discrete storage of the altitudes of a plot defined between four points. This plot is called a "slab." A slab is composed of a regular mesh where an altitude value is associated with each of the meshes. This altitude is generally that of a point on the digital model located at the center of the slab. The quality of a digital terrain model depends directly on the mesh interval. This interval is called resolution and defines the accuracy of the model. The smaller the resolution, the more accurate the digital terrain model.

[0004] The document "Review of Target Geo-Location Algorithms for Aerial Remote Sensing Cameras without Control Points" Yiming Cai et al. - Applied Sciences - 2022 - pp 1-27 discloses, in particular on page 9, a method for locating a target in a terrain using a digital terrain model. The method consists of:

[0005] - estimate the geographic location of the target and extract target elevation information from the digital terrain model,

[0006] - initialize a possible altitude as the maximum possible altitude of the target,

[0007] - determine the longitude and latitude of the target in the Earth's ellipsoidal reference frame,

[0008] - find, from the digital terrain model, the geodetic altitude information of a point having as coordinates the possible altitude, the longitude and the latitude of the target,

[0009] - compare this geodetic altitude with the possible altitude; if the geodetic altitude is greater than the possible altitude, return this altitude as the result, otherwise, start again from the second step, taking as the new possible altitude the previous possible altitude reduced by a small value.

[0010] No information is provided in this document regarding the method for estimating the geographic location of the target in the digital terrain model.

[0011] Document US9714833B2 discloses a principle for determining geographic coordinates of a point of interest, from a 2D image captured from a high position by a downward-facing optical sensor, with projection of the camera position onto a digital terrain model. This document discusses tracing a path in the direction of the point of interest and the use of a collision detection algorithm. It further discloses the application of this algorithm to the detection of forest fires.

[0012] A disadvantage of the principle described in the prior document is that no concrete implementation details are known. Given the hardware constraints, in particular the computing power available for a reasonable weight and cost of the equipment, the provision of a concrete solution for locating a point of interest remains non-obvious for those skilled in the art and is the subject of the present invention.

[0013] Robert T. Collins ET AL: "A System for Video Surveillance and Monitoring: VSAM Final Report", Technical report CMU-RI-TR-00-12, Robotics Institute, Carnegie Mellon University, May 1, 2000 (2000-05-01). XP055342999, Retrieved from the Internet: URL:http: / www.ri.cmu.edu / pub_files / pub2 / collins_robert_2000_1 / collins_robert_2000_1.pdf, discloses an application of the Bresenham algorithm to traversing a digital terrain model to determine the point at which a boresight axis meets the terrain.

[0014] This disclosure links the accuracy of the calculation to the resolution of the digital terrain model, which determines the required computing power. This can be quite substantial, to the point of preventing calculations by mobile means. However, in a search for a point of interest as in the envisaged application of vegetation fire detection, accuracy does not necessarily require following the resolution of the digital terrain model.

[0015] Generally speaking, we speak more specifically of an altitude field to define a digital terrain model.

[0016] The invention aims in particular to propose a simple solution for locating a point of interest that requires neither significant on-board resources nor real-time processing. As soon as a point of interest is captured in the camera's field of vision, it can be located, even at a later time.

[0017] To this end, the invention relates to a method for locating a point of interest on a terrain from an image, by selecting a mesh of a digital terrain model comprising meshes paving said terrain, each mesh being associated with an altitude representative of the part of the terrain that it paves, comprising the following steps:

[0018] - obtaining an image of the terrain taken by an optical sensor;

[0019] - identification and localization of the point of interest in the image;

[0020] - obtaining the position of the optical sensor at the time of image capture in a terrestrial reference frame – for example, altitude, longitude, latitude –;

[0021] - determination of the orientation of a sighting axis pointing towards the point of interest;

[0022] - projection, onto the digital terrain model, of the terrestrial position of the optical sensor, according to a predetermined projection direction;

[0023] - projection, onto the digital terrain model, of the line of sight, according to the same predetermined projection direction;

[0024] - determination, by an implementation of the Bresenham algorithm, of a mesh path in the digital terrain model, including the projection of the terrestrial position of the optical sensor and following the projection of the line of sight and

[0025] - search for the path mesh whose associated altitude corresponds to the distance between said mesh and the projection of said mesh on the line of sight, according to the predetermined projection direction.

[0026] The invention is characterized in that each mesh is assimilated to the point which is located in its middle and, according to an approximation, it is considered that the middles of the meshes are aligned on the orthogonal projection of the line of sight.

[0027] This approximation allows the altitude to be calculated by taking into account simple trigonometric considerations, instead of exactly following the geometry resulting from the projection of the line of sight onto the digital terrain model.

[0028] The mesh meeting the search criterion is the mesh containing the projection of the point of interest on the digital terrain model.

[0029] Determining the mesh path and searching for the mesh that meets the search criterion are two steps that can be intertwined. For example, it is possible that the mesh path is not fully determined but that only a first mesh of the path has been determined by implementing the Bresenham algorithm and that the search begins on this first mesh. If the mesh does not meet the criterion, the path determination can resume to determine a second candidate mesh, and so on.

[0030] Another advantage of the method according to the invention is that it can be implemented using a single optical sensor.

[0031] Yet another advantage of the method according to the invention is that it can be implemented after a single shot and that the processing of the image thus obtained can be carried out later. It is therefore not necessary to carry heavy or powerful computing resources on the flying machine carrying the camera.

[0032] According to a particular embodiment of the localization method, the meshes of the digital terrain model are gathered into slabs and the projections onto the digital terrain model are determined by:

[0033] - the slab containing the projection and

[0034] - the coordinates of said projection in a reference frame of said slab.

[0035] An advantage of using slabs is a simplification of the calculations required for each iteration. Determining the slab containing the projection is simply a comparison of the coordinates of the projected point with the coordinates of the four vertices of the square formed by the slab. Once this slab is determined, iterating from mesh to mesh is a very simple operation, along the mesh path determined by implementing Bresenham's algorithm inside the slab. It does not require large calculations.

[0036] According to a particular embodiment of the localization method, the iteration is stopped if the mesh considered in the iteration loop is the last of the mesh path located inside the slab.

[0037] In fact, if the slab exits, the algorithm must be repeated inside the neighboring slab in the direction of the line of sight.

[0038] According to a particular embodiment of the localization method, the digital terrain model is replaced by a digital surface model.

[0039] Although such an option is not preferred for a perfect implementation of the invention, it is within the scope of the invention.

[0040] According to a particular embodiment of the location method, the optical sensor is a camera operating in the visible and infrared spectra.

[0041] According to a particular embodiment of the location method, the predetermined projection direction is vertical.

[0042] Vertical projection is the most natural projection, which makes projection distance and altitude coincide. However, a non-vertical projection would not prevent the present invention from providing the expected result.

[0043] According to a particular embodiment of the location method, the expression "corresponds to" means "greater than or equal to", so that the search criterion for the path mesh whose associated altitude corresponds, according to the invention, to the distance between said mesh and the projection of said mesh on the line of sight, according to the predetermined projection direction, means that the altitude of the mesh is greater than or equal to the distance considered.

[0044] An advantage of this simple comparison criterion is that it is sufficient to compare the altitude associated with each cell in the digital terrain model with the distance between said cell and its projection on the line of sight. If the altitude of a cell is greater than or equal to said distance, this means that the line of sight has encountered the terrain, which reveals that the cell considered contains the point of interest. Thus, the point of interest is located in the digital terrain model.

[0045] According to a particular embodiment of the localization method, the search consists of an iteration from mesh to mesh, starting from the mesh containing the projection of the terrestrial position of the optical sensor, moving away from it in the direction of the point of interest.

[0046] According to a particular embodiment of the localization method, the iteration is stopped if a predetermined number of iteration loops is exceeded.

[0047] This limit avoids an infinite loop if, for some reason, the search criterion for the mesh containing the point of interest was never satisfied.

[0048] According to a particular embodiment of the location method, the representative altitude is an average altitude.

[0049] In a particular embodiment, the point of interest is a hot spot in the image, which makes it possible to detect the start of vegetation fires.

[0050] Another object of the invention is a device for locating a point of interest on a terrain from an image, characterized in that it comprises a computer-readable memory medium, with instructions encoded thereon, and at least one processor connected to the memory medium and configured to read and execute the instructions and, when it executes the instructions, implement the operations of the method. Brief description of the figures

[0051] The invention will be better understood on reading the following description, given solely by way of example and with reference to the appended drawings in which:

[0052] is a schematic view of the different mathematical references used to implement the invention,

[0053] is a diagram illustrating the implementation of Bresenham's algorithm,

[0054] represents a plane of the Bresenham reference frame cut into octants,

[0055] provides an excerpt of a simple logic code of an execution loop of Bresenham's algorithm,

[0056] represents a decision tree translating the execution of the code of the,

[0057] provides an excerpt from a refined logic code of an execution loop of Bresenham's algorithm. Detailed description

[0058] We will now describe how a point of interest can be located in an image on a terrain.

[0059] Localization is performed in two stages: detection and localization of the point of interest in the image and geographical localization of the point of interest on the ground.

[0060] A camera 6 on board a drone 7 is an optical sensor that provides an image 8 of the monitored area. In this image 8, the point of interest constitutes a pixel 9 having its own coordinates in this image 8. Different detection algorithms can be implemented to detect the point of interest. For example, if it is a hot spot in an image for fire detection, the camera can be an infrared camera and the temperature of each pixel in the image is measured so as to determine the hot spot which then constitutes the point of interest.

[0061] Once this point of interest has been determined, the geographical location of this point is obtained by the ray-tracing method, according to which a line of sight 10 going from the camera to the point of interest is considered. This line of sight 10 has an orientation provided by a vector. The camera 6 has coordinates. The combination of the information consisting of the coordinates of the camera 6 and the orientation of the vector makes it possible to find the geographical coordinates of the hot spot according to the method which will be described.

[0062] First of all, it is a matter of carrying out mathematical transformations to move from the camera 6 reference point to the terrestrial reference point.

[0063] This succession of transformations of reference points requires the following reference points.

[0064] - Land marker “East North Rising” or “ENU” 101, R M ENU, in which the X axis of the abscissas looks towards the east, the Y axis of the ordinates looks towards the north and the vertical Z axis looks upwards. The origin point is located at a reference point of a digital terrain model 12. This reference point is assimilated to the reference mesh of the digital terrain model.

[0065] - Land marker “North East Descending” or “NED” 102, R M NED , in which the X axis looks north, the Y axis looks east, and the Z axis looks down. The origin of the coordinate system is placed at the center of gravity of drone 7. This coordinate system is mainly used to determine the orientations of drone 7.

[0066] - Bresenham Landmark, R B 103, oriented as the East North Rising R marker M ENU , but whose origin is placed at the starting point of the algorithm which will be described.

[0067] - Drone reference, R D104. This reference point is placed at the center of gravity of drone 7 with the X axis looking forward, the Y axis looking to the right and the Z axis looking down. The angles between drone 7 and the North East Descending reference point are called the pitch of the drone (D p ), the roll of the drone (D r ) and the drone's yaw (D y ).

[0068] - Cardan reference, R G 105. This reference point is oriented and placed like the drone's reference point. The angles between the drone 7 and the camera 6 are called camera pitch (C p ), camera roll (C r ), and camera yaw (C y ).

[0069] - Camera reference, R C 106, which is placed at the optical center of camera 6, with the Z axis looking forward along the optical axis, the Y axis looking down, and the X axis looking to the right.

[0070] - Image reference, R I107. This marker is placed in the center of image 8 with the X axis looking to the right, the Y axis looking down, and the Z axis pointing into the image.

[0071] - Pixel marker, R P 108. This marker is oriented like the marker in image 7, but its origin is placed at the upper left corner of image 8.

[0072] The distance between the origins of the drone reference frame 104 and the gimbal reference frame 105, as well as the distance between the origins of the gimbal reference frames 105 and camera 106, are negligible. Thus, it is considered that the origins of the terrestrial reference frames NED R M NED 102, drone R D 104, cardan R G 105 and R camera C 106, are confused.

[0073] La represents each of the previously explained reference points, as well as the successive coordinate transformation operations to move from the reference point of the image R I 107, in image 8, at the Bresenham R marker B103, in the digital terrain model 12, passing successively through the pixel reference point R P 108, the R camera marker C 106 on camera 6, the gimbal mark R G 105, the drone R's landmark D 104 on drone 7, terrestrial reference point R M NED 102, the terrestrial reference point R M ENU 101.

[0074] To geolocate the point of interest in the image, we proceed in two steps. First, the direction vector 11i of the line of sight 10 is determined. Its coordinates are transformed to pass from the pixel reference R P 108 at the terrestrial reference point R M ENU 101; Then, the Bresenham algorithm will be used to traverse the digital terrain model and find the location of the point of interest, in accordance with the invention.

[0075] The geolocation algorithm starts with pixel 9 of the point of interest identified in image 8. The coordinates of this pixel 9 are (u, v, 1) T and the vector 11p composed of its coordinates is named T p . Five transformations are required to transform the coordinates of this point of interest, expressed in reference point 107 of image 8, into coordinates expressed in the terrestrial reference point “East North Upward”, R M ENU 101.

[0076] First transformation 1: from the pixel reference R P 107 to camera marker R C 106 of camera 6. This transformation uses the intrinsic matrix of camera 6. For a theoretically perfect camera, the intrinsic matrix takes the following form.

[0077]

[0078] in which f x and f y are respectively the focal length on the u and v axes of the pixel frame and C x etc ythe coordinates of the point of interest in the image in pixels. The transformation of the vector T p in the pixel reference frame towards T c , designated 11c, in the camera 6 frame is

[0079]

[0080] Second transformation 2: from the camera reference point R C 106 towards the R cardan mark G 105. This transformation is necessary to align the camera coordinate system and the gimbal coordinate system. It requires two rotations, one around the Z axis G from the gimbal reference point, the other around the Y axis G of the gimbal reference, each of +90°. These two rotations are summarized but in the following matrix.

[0081]

[0082] to transform the vector T c from the camera reference to the vector T g from the gimbal reference point, we apply the formula

[0083]

[0084] Third transformation 3: from the gimbal reference point R G 105 to drone marker R D 104 of drone 7. This transformation takes into account the orientation angles of the camera 6 C y , C p etc r . As illustrated in the, the rotational matrices between the gimbal frame R G 105 and the drone marker R D 104 can be calculated as follows:

[0085]

[0086]

[0087]

[0088]

[0089] So, to transform the vector T g expressed in the gimbal frame of reference R G 105 into a vector T d expressed in the drone reference R D 104, the formula to be applied is as follows.

[0090]

[0091] Fourth transformation 4: from drone reference R D 104 towards the terrestrial marker “North East Descending” RM NED 102. This transformation takes into account the orientation angles of the drone D y , D p and D r . From there, the rotational matrices between the drone frame R D 104 and the terrestrial reference point R M NED 102 can be calculated. The formats of each of these matrices are the same as for the third transformation, namely:

[0092]

[0093] Thus, to transform the vector Td expressed in the drone frame R D into a vector T m ned expressed in the terrestrial reference frame R M NED , the formula is as follows.

[0094]

[0095] Fifth transformation 5: from the terrestrial reference point R MNED 102 towards the terrestrial reference point R M ENU101 in the digital terrain model 12. This transformation is necessary due to the orientation angles of the drone defined according to the terrestrial reference R M NED 102. Since the digital terrain model 12 is defined in the terrestrial reference frame R M ENU 101, the transformation of the R reference frame M NED 102 towards the R marker M ENU 101 requires a rotation of +90° around the z axis menu , then another rotation of +180° around the y axis menu . These rotations can be summarized by the following matrix.

[0096]

[0097] Finally, to transform the vector Tmned of the terrestrial reference frame R M NED 102 in a vector Tmenu = T, designated 11, in the terrestrial reference frame ENU 101, the formula is as follows

[0098]

[0099] We will now determine the altitude of the line of sight 10. Once the direction vector 11 of the line of sight is expressed in the terrestrial reference frame R M ENU 101, the Bresenham algorithm can be used to traverse the digital terrain model 12 and find the location of the point of interest. Lamontre shows this route.

[0100] The route starts at mesh 13 (x dep , y dep ) of the digital terrain model which corresponds to the location of the drone, by projection according to a predetermined projection direction, here the vertical. The drone is located on the Z axis at a certain altitude (Drone Alti ).

[0101] We set a maximum number of iterations k to execute the algorithm.

[0102] The algorithm consists of a succession of iterations, with a maximum number of k, during which the altitude deduced from the line of sight is compared to the altitude provided by the digital terrain model. As illustrated by the, the digital terrain model is run over a succession of cells 13 to 18. Each cell is assimilated to the point located in its middle and, according to an approximation considered acceptable, although geometrically false, these midpoints are considered to be aligned with the orthogonal projection of the line of sight. At each iteration, the altitude is calculated by taking into account simple trigonometric considerations. If we consider in a simplified way the triangle ABC on the, to find the altitude of the line of sight, we perform the product of the vectors T C and N or N points down and is the normal vector of the x-plane b y bThe purpose of this scalar product is to find the delta angle, which we call the "propagation angle":

[0103]

[0104] with And so, for the inner product, as a term-by-term product:

[0105]

[0106] Or And are standardized:

[0107] Once the propagation angle is calculated, we have:

[0108] Or

[0109] Thus, the altitude of the line of sight (Altiray) is:

[0110]

[0111] For each mesh, the altitude of the line of sight is the distance between said mesh and the projection of said mesh on the line of sight, according to the projection direction.

[0112] Knowing the two altitudes at each iteration, the comparison can be made.

[0113] If the altitude of the line of sight 10 is less than or equal to the altitude provided by the digital terrain model 12, then this means that the line of sight 10 intercepts the Earth's surface and that the point of interest is located in the cell of the digital terrain model 12 considered at this iteration. Otherwise, the Bresenham algorithm provides a next cell to explore and the comparison is carried out again on this next cell, and so on until the line of sight 10 intercepts the digital terrain model 12. When this interception occurs, the altitude of the line of sight 10, that is to say the distance between said cell and its projection on the line of sight, according to the projection direction, is greater than the altitude associated with the cell in the digital terrain model.

[0114] We will now provide a more detailed description of Bresenham's algorithm.

[0115] The Bresenham algorithm is an algorithm for tracing a straight line in a mesh of dimension n, by determining which mesh cells must be retained to form said straight line. In the case of the invention, the dimension n is equal to 2.

[0116] As shown in the, the x-plane b y b is separated into eight parts 19 to 26 designated "octants", based on the inclination of the vector. In this way, the method can be performed for the first octant 19 only and extended to the other octants 20-26 using the following symmetries:

[0117] • invert the x and y components, which amounts to going from octant 19 to octant 20,

[0118] • change the sign of the x and y components, which amounts to going from octant 22 to octant 26.

[0119] Laprovides an excerpt from a logic code of a loop executing Bresenham's algorithm. From a starting point with coordinates (x0, y0), a vector (dx, dy) is drawn on the first octant 19. The coordinates dx and dy refer to the inclination of the vector on the x-axis and on the y-axis respectively.

[0120] Considering a pixel with coordinates (x, y), we calculate a decision variable named “d” as follows:

[0121]

[0122] in order to determine which next pixel should be chosen from the following possibilities: (x +1, y) and (x+1, y+1), depending on the sign of the value d. If d is negative, then the next pixel will be (x +1, y). Otherwise, the next pixel will be (x+1, y+1).

[0123] The errors accumulated during mesh exploration are taken into account. When crossing a pixel with coordinates (x+1, y), the error is εx and if the pixel with coordinates (x+1, y+1) is crossed, the error is εxy.

[0124]

[0125]

[0126] By running the algorithm, we follow a decision tree represented in. Four situations can arise:

[0127] - if dx = 0, only y is incremented,

[0128] - if dy = 0, only x is incremented,

[0129] - if dx = dy, the exploration must be done on the diagonal, which means that both x and y must be incremented.

[0130] A differentiation must be made between the eight octants of the plane.

[0131] - if dx > dy, the bytes considered are octants 19, 22, 23 and 26. In this case, the propagation can directly follow the sign of the vector components. The signs of the components are stored in two specific variables sx and sy.

[0132] - if dx < dy, the octants considered are 20, 21, 24 and 25. Thanks to the symmetrical properties, we return to the case in which dx is greater than dy, by playing on the symmetries. For this, we must consider two sub-cases:

[0133] - if (dx > 0 and dy > 0) or (dx < 0 and dy < 0), i.e. dx and dy have the same sign, the octants processed are 20 and 24 and the x and y coordinates must be reversed

[0134] - if (dx > 0 and dy < 0) or (dx < 0 and dy > 0), that is to say dx and dy have opposite signs, the octants considered are 21 and 25. The x and y coordinates must be reversed, then their signs changed.

[0135] An offset variable is introduced to account for the fact that the center of the x-plane b y b is not always zero. In this case, the algorithm can be refined according to the logic code presented in.

[0136] Other, less efficient algorithms can also be used instead of the Bresenham algorithm, such as the midpoint algorithm or the DDA (for "Digital Differential Analyzer") algorithm. An advantage of the Bresenham algorithm in the context of the invention is that the digital terrain model can be considered as a large screen whose pixels are the meshes of the digital model.

[0137] In summary, the steps that are performed for the execution of Bresenham's algorithm are as follows:

[0138] (a) Initialization of the variable k = 0.

[0139] (b) At the coordinates (x, y) of point O, projection onto the digital terrain model 12 of the position of drone 6, the corresponding height h is sought in the digital terrain model.

[0140] (c) The value h is then compared with the altitude of drone 6, taken as the initial value of the variable H.

[0141] (d) As H > h, k is incremented by 1 and the coordinates of the next point are calculated.

[0142] (e) This process is repeated as many times as necessary, giving the variable H the altitude of the next point and comparing H to h.

[0143] (f) When the condition H < h is satisfied, the point is identified.

[0144] This point corresponds to the intersection between the mesh path and the Earth's surface.

[0145] Alternatively, when the mesh path propagates on the digital terrain model, it may be that for some reason or other, the stopping condition H < h is never satisfied. Two additional stopping conditions can be introduced to overcome this problem. The first condition concerns a maximum number of iterations to be performed for propagation. The second condition makes the program stop when the mesh path leaves the slab of the digital terrain model. This last condition will also make it possible to know, if the starting point is in one slab and the desired point in another, at which point we leave the first slab and by extrapolation to know at which point we enter the second. Thus, we can travel the digital terrain model in all possible directions.

[0146] The device used for the present implementation of the invention comprises four entities (not shown):

[0147] - the first, in the form of an Android application, corresponding to an interface between the drone and the remote pilot. The latter can choose to carry out a flight in manual piloting or to create a trajectory for an automatic mission. This automatic mission is generated on the basis of any polygon made up of at least three points entered on a satellite map. || there are two modes for generating this trajectory from the flight height, the coverage rate as well as the camera parameters: scanning the surface of the polygon; and tracing the contour of the polygon, with an additional parameter: the width to be scanned entered by the user. If the requested width is greater than that possible at this height, the trajectory generation will then automatically create one or more polygons, uniformly dilated, in order to fly over an area around the polygon in relation to the requested width.These trajectory generation strategies are adapted to the needs of firefighting operators for monitoring burned areas and fire edges. Once the trajectory has been validated by the remote pilot, the mission is then generated based on the drone's performance and the speed entered by the remote pilot. This mission is transferred to the drone. During the flight, the application will allow these essential parameters to be monitored and its route to be tracked.

[0148] - the second, which can be developed in C++, retrieves the video stream from the thermal camera and processes it to automatically detect hot spots based on a fixed temperature threshold. Once detected, each hot spot is assigned its pixel coordinates, the GPS position of the drone as well as the orientation of the camera at the time they were taken.

[0149] - the third entity, which can be developed in Python, retrieves the attributes of each hotspot to determine its geographic coordinates on the Earth's surface. This operation is performed by tracing a ray propagating on a digital terrain model via a Bresenham algorithm;

[0150] - finally, in a fourth and final entity, the data collected during the flight are archived on the on-board computer and transmitted to the remote pilot without a wired connection.

[0151] The steps of the process can thus be summarized as follows:

[0152] - projection of the image capture position onto the digital terrain model,

[0153] - determination of the direction in which the point of interest is located on the digital terrain model,

[0154] - from the projected position, tracing of a path of meshes in the determined direction, each mesh of the path being determined using the Bresenham algorithm adapted to vectorization, with, at each mesh of the traced path, the comparison between the altitude of the corresponding geographical point and that of the mesh,

[0155] - the process is stopped if the altitudes cross.

[0156] The result of this process is the reference of the mesh 18 corresponding to the point of interest. List of references

[0157] 1 to 5: Change of reference frame transformations

[0158] 6: camera

[0159] 7: drone

[0160] 8: image

[0161] 9: pixel

[0162] 10: sight axis

[0163] 11, 11c, 11i, 11p: direction vector

[0164] 12: digital terrain model

[0165] 13-18: meshes of the digital terrain model

[0166] 19-26: octants of the x-plane b y b

[0167] 101 = R M ENU : Land reference “East North Rising” or “ENU”

[0168] 102 = R M NED : Terrestrial reference point “North East Descending” or “NED”

[0169] 103 = R B : Bresenham Landmark

[0170] 104 = R D : Drone reference

[0171] 105 = R G : Cardan reference

[0172] 106 = R C : Camera reference

[0173] 107 = R I : Image reference

[0174] 108 = R P : Pixel marker

Claims

Method for locating a point of interest (9) on a terrain from an image (8), by selecting a mesh of a digital terrain model (12) comprising meshes (13-18) paving said terrain, each mesh being associated with an altitude representative of the part of the terrain that it paves, comprising the following steps: - obtaining an image (8) of the terrain taken by an optical sensor (6); - identifying and locating the point of interest (9) in the image (8); - obtaining the position of the optical sensor (6) at the time of image taking in a terrestrial reference frame (101); - determining the orientation of a line of sight (10) pointing towards the point of interest (9); - projecting, onto the digital terrain model (12), the terrestrial position of the optical sensor (6), according to a predetermined projection direction; - projecting, onto the digital terrain model (12), the line of sight (10), according to the same projection direction predetermined;- determination, by an implementation of the Bresenham algorithm, of a path of meshes (13-18) in the digital terrain model (12), including the projection of the terrestrial position of the optical sensor (6) and following the projection of the line of sight (10) and - search for the mesh of the path whose associated altitude corresponds to the distance between said mesh and the projection of said mesh on the line of sight (10), according to the predetermined projection direction, characterized in that each mesh is assimilated to the point which is located in its middle and, according to an approximation, it is considered that the middles of the meshes are aligned on the orthogonal projection of the line of sight.; Location method according to claim 1, in which the meshes of the digital terrain model are grouped into slabs and the projections onto the digital terrain model (12) are determined by: - ​​the slab containing the projection and - the coordinates of said projection in a reference frame of said slab. A localization method according to claim 2, wherein the iteration is stopped if the mesh considered in the iteration loop is the last of the mesh path located inside the slab. A location method according to any one of claims 1, 2 and 3, wherein the digital terrain model is replaced by a digital surface model. A location method according to any one of claims 1, 2, 3 and 4, wherein the optical sensor is a camera (6) operating in the visible and infrared spectra. A locating method according to any one of claims 1, 2, 3, 4 and 5, wherein the predetermined projection direction is vertical. A method of locating according to any one of claims 1, 2, 3, 4, 5 and 6, wherein the expression "corresponds to" means "greater than or equal to". Location method according to any one of claims 1, 2, 3, 4, 5, 6 and 7, in which the search consists of an iteration from mesh to mesh, starting from the mesh containing the projection of the terrestrial position of the optical sensor, moving away from it in the direction of the point of interest. A localization method according to claim 8, wherein the iteration is stopped if a predetermined number of iteration loops is exceeded. A location method according to any one of claims 1, 2, 3, 4, 5, 6, 7, 8 and 9, wherein the representative altitude is an average altitude. A localization method according to any one of claims 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10, wherein the point of interest is a hot spot in the image. Device for locating a point of interest on a terrain from an image, characterized in that it comprises a computer-readable memory medium, with instructions encoded thereon, and at least one processor connected to the memory medium and configured to read and execute the instructions and, when executing the instructions, implement the operations of the method according to any one of claims 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10.

Images