system and method for constructing a map of an environment
By preselecting initial points and applying local corrections to the measuring device's movement, the system accurately identifies areas of interest, improving map precision and reducing calculation times in SLAM techniques.
Patent Information
- Application Number
- FR2022011164
- Authority / Receiving Office
- FR · FR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2022-10-26
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-10-26
AI Technical Summary
Existing SLAM techniques struggle to accurately identify areas of interest in an environment for constructing a precise map due to subjective and imprecise thickness criteria, leading to blurry or inaccurate maps when determining the trajectory of the measuring device.
A system and method that preselects sets of initial points, applies local corrections to the measuring device's movement based on potential areas of interest, and jointly corrects the trajectory to ensure that only actual areas of interest are included in the map construction, using criteria such as minimum thickness, alignment with a plane, or cylindrical shape.
This approach allows for more precise identification of areas of interest, ensuring a clear and accurate map construction by iteratively correcting the trajectory and speed of the measuring device, enabling real-time operation and reducing calculation times.
Smart Images

Figure 00000019_0000 
Figure 00000019_0001
Abstract
Description
Title of the invention: System and method for constructing a map of an environment FIELD OF INVENTION
[0001] The present invention relates to a system and a method for constructing a map of an environment. More specifically, the invention relates to the field of automatic mapping of an environment from a three-dimensional point cloud acquired by a mapping device comprising a lidar (from the English acronym Light Detection And Ranging). technological BACKGROUND
[0002] In the field of automatic mapping, an environment can be mapped using a measuring device, mobile within the environment, comprising at least one lidar. The lidar scans the environment in which the device moves and thus acquires point clouds representative of the environment in which it operates. Such an environment can be a building, a road, a construction site, etc. The map of the environment is built as points are acquired by the lidar, i.e., as the measuring device comprising the lidar moves.
[0003] However, in some cases, the resulting environmental map may not be representative of the environment, at least in certain areas. In this case, the environmental map will appear blurry. This can occur because the lidar takes relative measurements; therefore, it is necessary to know its position and orientation in order to locate the points it returns with each acquisition within a global orthogonal coordinate system that situates the mapped environment.
[0004] When lidar measurements are used to jointly calculate the position and orientation of the device and the map, the technique implemented is called the SLAM technique (for the English acronym Simultaneous Localization and Mapping). It consists of constructing the map of the environment using a measurement device including a lidar, combined with identifying the trajectory and orientation of the measurement device.
[0005] In particular, the trajectory of the measuring device is not known at every instant. Thus, each hypothetical trajectory of the measuring device can be associated with a different map of the environment. The quality of the environmental map is notably conditioned by knowledge of the trajectory of the The measurement device requires finding the correct trajectory within the environment to obtain a high-quality environmental map. Consequently, the device's trajectory can be identified by searching for the one that leads to the best quality environmental map.
[0006] The search for the trajectory and the best quality map is carried out simultaneously.
[0007] To achieve this, it is possible to define areas of interest in the environment. Such areas may include, in particular, a flat area of the environment, for example, a wall, a facade, a road, etc. Such areas of interest have the advantage of being easily mapped, and are therefore useful for estimating the trajectory of the measuring device in the environment. The aim is then to focus on such areas to improve the trajectory estimation and obtain a high-quality map.
[0008] One of the challenges of SLAM is that mapping tends to be done automatically, without an external person defining which areas are of interest for estimating the trajectory of the measuring device.
[0009] In this perspective, the prior art proposes to identify approximately flat areas, considered as areas of interest, on the point cloud and to select some of these areas as flat areas according to a quality criterion corresponding to the thickness of these areas.
[0010] More specifically, the known method consists of performing a first approximate reconstruction of a point cloud from a prior device trajectory. Approximately flat areas on the point cloud are identified. Their thickness is evaluated to determine if it corresponds to a predefined thickness. This involves comparing the thickness of the point cloud along an area with the predefined thickness. If the thickness of the point cloud along an area corresponds to the predefined thickness, these areas are selected as flat, and then the overall device trajectory is modified to reduce the thickness of the approximately flat areas found, that is, to make them effectively flat.
[0011] However, this solution is not satisfactory. Indeed, the thickness criterion is subjective and imprecise. For example, if an excessively large predefined thickness is tolerated, there is a risk of selecting areas that are not actually flat.
[0012] On the other hand, if a predefined thickness that is too low is tolerated, there is a risk of finding very few flat areas. This risk arises from the fact that, for a predefined thickness that is too low, the trajectory of the measuring device must be known with a correspondingly high degree of precision. Indeed, approximately flat areas only appear truly flat once the trajectory of the measuring device is known. However, when comparing the approximately flat area with With the predefined thickness, the only available environmental map is an approximate reconstruction of the environment based on an approximate trajectory of the measuring device. Therefore, if the predefined thickness is too thin, areas that are actually flat may not meet this criterion on the approximate map. Furthermore, when such areas do meet the thickness criterion, it means they are already very thin. Consequently, they cannot be used to make subsequent corrections to the trajectory of the measuring device.
[0013] The invention aims to solve at least partially the technical problems outlined. Summary of the invention
[0014] Thus, the invention relates to a system for constructing a map of an environment, the system comprising: - a measuring device comprising at least one lidar configured to acquire a point cloud representative of said environment in which the measuring device moves, - a processor configured for * perform a preselection of a plurality of initial point sets in the point cloud, each initial point set in the plurality of initial point sets being representative of a potential area of interest in the environment, *For each initial set of points, find a local correction to apply to a movement of the measuring device along each potential area of interest represented by the initial set of points considered in order to determine if each initial set of points actually represents an area of interest. *Jointly correct the initial point sets effectively representing an area of interest, by applying a global correction to a movement of the measuring device in the environment, Construct the environmental map with the corrected initial point sets effectively representing an area of interest and the initial point sets not representing an area of interest.
[0015] The invention also relates to a method for constructing a map of an environment, comprising: - The acquisition, by a measuring device comprising at least one lidar, of a point cloud representative of said environment in which the measuring device moves, - Implementation by a processor: *From a preselection of a plurality of initial point sets in the point cloud, each initial point set of the plurality of initial point sets being representative of a potential area of interest in the environment, *for each initial set of points, a search for a local correction to be applied to a movement of the measuring device along each potential area of interest represented by the initial set of points considered in order to determine if each initial set of points actually represents an area of interest, *a joint correction of the initial sets of points actually representing an area of interest, by applying a global correction to a movement of the measuring device in the environment, *a construction of the environment map with the corrected initial point sets effectively representing an area of interest and the initial point sets not representing an area of interest.
[0016] Thus, contrary to what is known in the prior art, areas of interest are not identified by looking at the approximately constructed map to locate areas resembling a plane, but by performing local trajectory corrections on sets of points. These corrections aim to transform certain areas into areas of interest, thereby determining whether a set of points actually represents an area of interest or not. Advantageously, all points in the point cloud belong to an initial set of points. Thus, the system makes it possible to efficiently determine which potential area of interest is indeed an area of interest: in fact, if a local trajectory correction along a potential area of interest reveals that it is indeed an area of interest, this means that the potential area of interest is indeed an area of interest.The overall trajectory correction can therefore be based on the area of interest actually determined.
[0017] In particular, when the processor attempts to correct the trajectory locally, the local correction may reveal a region of interest, or the applied local correction may reveal that the set of points under consideration is not a region of interest. For example, if the region of interest sought is a flat area, the local correction of the trajectory along the set of points under consideration will make the flat area appear. Conversely, if the set of points under consideration is not a flat area, no local correction will make a flat area appear.
[0018] Depending on different aspects, it is possible to foresee one and / or the other of the characteristics below taken alone or in combination.
[0019] According to one embodiment, the processor is configured to iterate through the preselection, local correction and joint correction steps until a predefined criterion is met.
[0020] This arrangement allows for a more precise correction of the local trajectory, and / or for determining more certainly whether an area of interest is indeed an area of interest or not.
[0021] According to one embodiment, the searches for local correction to be applied to each initial set of points are carried out in parallel.
[0022] Thus, areas of interest are identified more quickly than if the processor were to search for local corrections sequentially. This allows for a system that addresses the SLAM problem and operates in real time and instantaneously or almost instantaneously.
[0023] According to one embodiment, the local correction to be made to a movement of the device includes the correction of a local trajectory of the device along each potential area of interest and in which the joint correction includes the correction of a global trajectory of the measuring device in the environment.
[0024] According to one embodiment, the local correction to be made to a movement of the device includes the correction of a local velocity of the device along each potential area of interest and in which the joint correction includes the correction of an overall trajectory of the measuring device in the environment.
[0025] The correction can therefore be made to the trajectory or to the speed. The two corrections can be made in parallel or can be interchangeable.
[0026] According to one embodiment, the overall correction to be applied to a movement of the measuring device in the environment is determined by solving an optimization problem.
[0027] According to one embodiment, the areas of interest include a flat area such as a road, a signpost, or a wall, the processor being configured, for each potential area of interest corresponding to a potential flat area, to calculate a minimum thickness C' of said potential flat area obtained after local correction of the movement of the measuring device, said minimum thickness being obtained by direct resolution of: C(XT) = A(^[£XvLLrA']^® 7 ')' où N is the number of points xi in the initial set of points considered, 4' — is the average of the points xi in the initial set of points, t — is the average of the times associated with said points in the initial set of points, _ v is an intermediate calculation, S~~^Ÿ _ ÿ) t is the covariance matrix, and The function A(■) returns the smallest eigenvalue of the matrix given as an argument. and where the processor is configured to determine that the potential flat area is indeed an area of interest if C' is less than a predefined threshold.
[0028] Thus, the calculation of the criterion for determining whether an area is indeed an area of interest can be done directly. It does not require an intermediate optimization step. This advantage is specific to the case where the correction is applied to the speed and when the area of interest is a flat area.
[0029] According to one embodiment, the areas of interest include a cylindrical area such as a post, the processor being configured, for each potential area of interest corresponding to a potential cylindrical area, to determine a correction of a local movement (Amouv) of the measuring device for which the distance is minimal between each of the N points xi of the initial set of points considered and a cylinder of center c, axis u and radius r according to a double minimization: ^min min AL, | \ / H Amouv) - c 112 - [ | (x,(Amouv) - c) | p -r| where Amouv is a variation of the motion of the measuring device 10 at each point Xi, and where xi=i>2, ...(Amouv) is the position of point i after application of a variation of motion Amouv, the processor is further configured to determine that a potential cylindrical area of interest is indeed a cylindrical area of interest if a result of the double minimization is less than a predefined threshold.
[0030] Thus, the processor is configured to identify areas of interest corresponding to cylindrical areas.
[0031] According to one embodiment, for each initial set of points effectively representing an area of interest, the processor is configured to remove the points present in the initial set of points but not belonging to said area of interest.
[0032] This avoids distorting the overall trajectory correction by considering points that are not part of an area of interest. Brief description of the drawings
[0033] Embodiments of the invention will be described below with reference to the drawings, briefly described below:
[0034] [Fig-1] represents a mapping system according to one embodiment.
[0035] [Fig.2] illustrates steps of a process for constructing a map of an environment using the system according to an embodiment.
[0036] [Fig.3] illustrates steps of a process for constructing a map of an environment using the system according to another embodiment.
[0037] In the drawings, identical references designate identical or similar objects. DETAILED DESCRIPTION
[0038] The invention relates to a mapping system shown in [Fig. 1]. The system 1 comprises a measuring device 10 including at least one lidar 101. The measuring device may also include an inertial measurement unit (IMU) 102 and a camera 103. The IMU may include at least one accelerometer and at least one gyroscope.
[0039] The measuring device can be mounted on a mobile object (not shown). The object can be a drone, a backpack worn by a moving person, a car, a robot, or any other mobile object. The mobile object comprising the measuring device 10 moves within an environment to be mapped. Such an environment can be indoors, for example inside a building, or outdoors, for example outside a building, on a street, on a road, at a construction site, etc.
[0040] As the object moves, the lidar is configured to perform measurements. These measurements include the acquisition of point clouds representative of the environment. From the point cloud, at least, a map of the environment can be constructed.
[0041] The system 1 further comprises a processor 11. The processor 11 can be connected to the measuring device so as to receive the data measured by the lidar, the camera, and / or the IMU. In one configuration, the communication between the processor 11 and the measuring device 10 can be wired. In another configuration, the communication between the processor 11 and the measuring device 10 can be wireless, for example, via radio waves, Bluetooth, or Wi-Fi.
[0042] In one configuration, the measuring device 10 and the processor 11 exchange data via a respective communication interface 104 and 110.
[0043] In one configuration, the processor 11 can be embedded in the measuring device 10. In another configuration, the processor 11 can be hosted on a remote server. According to this configuration, the communication is therefore remote, for example by radio waves, passing for example through the 4G and / or 5G network, Bluetooth or wifi.
[0044] System 1 may also include graphics cards as well as one or more high-capacity hard drives, on the order of terabytes.
[0045] Fig. 2 illustrates steps in a process of constructing a map of an environment using system 1.
[0046] At step SI, the measuring device 10 acquires data.
[0047] More specifically, the lidar 101 acquires at least one point cloud representative of the environment. In a preferred configuration, the measuring device 10 moves at each instant and the lidar acquires a point cloud at each instant. Thus, a plurality of point clouds are acquired when the measuring device 10 moves moves within the environment.
[0048] The IMU can also acquire data, including data concerning the orientation and acceleration of the measuring device 10 as it moves through the environment.
[0049] The data acquired by the lidar and / or the IMU are used to construct the environmental map. More specifically, the environmental map can be constructed after a single pass of the measuring device through the environment. The point clouds acquired by the lidar and the orientation and acceleration data of the measuring device 10 acquired by the IMU during a single pass of the measuring device 10 through the environment are used to construct the environmental map. More precisely, the IMU's acceleration measurement, which is called the specific force, is defined as the difference between the acceleration and gravity. This measurement indirectly provides information about the device's acceleration.
[0050] In another embodiment, the environmental map can be constructed after several passes of the measuring device through the environment. The point clouds acquired by the lidar and the orientation and acceleration data of the measuring device 10 acquired by the IMU during the passes of the measuring device 10 through the environment are used to construct the environmental map.
[0051] However, in order to construct a map of satisfactory quality, that is to say a clear and not blurry map, it is necessary to know the trajectory of the measuring device 10 as it passes through the environment. Indeed, when the trajectory of the measuring device is not known precisely, the map of the environment appears blurry.
[0052] The search for the trajectory of the measuring device and the construction of the map from the data acquired by the lidar 102 and the IMU 103 are carried out at the same time, and advantageously at the same time and in real time.
[0053] In order to determine the trajectory of the measuring device 10, it may be advantageous to define areas of interest along which the trajectory search can be facilitated. These areas of interest may be, for example, a flat area such as a wall, a roadside, a sidewalk, etc., or objects such as traffic signs and their supports, or even wall corners.
[0054] More specifically, each point in the point clouds acquired by the lidar 101 is described by three coordinates describing a position in the orthogonal frame attached to the measuring device, as well as a measurement time t. The points are also described by an intensity representing the fraction of the laser light energy emitted by the lidar 101 and reflected by the material in contact with it. The areas of interest can therefore be determined based on a criterion of point distribution in space (for planar areas, for example) or on an intensity criterion.
[0055] In order to identify areas of interest and construct a map representing cor- To precisely identify these areas of interest, the processor performs a preselection of a plurality of initial point sets in the point cloud at step S2, each initial point set of the plurality of initial point sets being representative of a potential area of interest in the environment.
[0056] The areas of interest are not known before the final map construction. Consequently, it is necessary to determine which areas of the environment are actually areas of interest. In order to miss as few areas of interest as possible that could help in constructing an accurate map of the environment, the initial point sets preselected by the processor can cover all the points in the point cloud. More precisely, each point in the point cloud is included in at least one initial point set. In other words, all areas of the environment can be preselected by the processor in step S2, in order to study them and determine whether they actually represent an area of interest, as described in step S3.
[0057] Then, in step S3, for each initial set of points, the processor 11 searches for a local correction to be made to a movement of the measuring device 10 along each potential area of interest represented by the initial set of points considered in order to determine if each initial set of points actually represents an area of interest.
[0058] For example, the areas of interest representing a flat area of the environment only appear flat, i.e. having a low thickness in the map of the environment, once the true trajectory of the measuring device 10 along this flat area has been determined.
[0059] For areas of interest representing a signpost, or a corner of a wall, for example, they only appear clear once the true trajectory of the measuring device 10 along this area of interest has been determined.
[0060] In the case of areas of interest representing a planar surface, in order to determine whether potential areas of interest are indeed planar, the processor 11 will locally correct a movement of the measuring device corresponding to a velocity of the measuring device 10 along each potential area of interest. The local velocity correction is independent for each potential area of interest, so that the processor 11 can search for each local correction in parallel. This significantly reduces computation time, so that the search for the trajectory of the measuring device 10 can be performed in real time.
[0061] Mathematically, the search for the local correction to be applied to the movement of the measuring device 10 along each potential zone of interest includes the definition of a criterion C (X, T) applied to each initial set of points X = (Xj, x2, - -. ) representing the potential zone of interest and to a series of instants associated T = ( tf?, ... ) to the points of the initial set of points, and where C' corresponds to the smallest quadratic thickness, i.e. the minimum squared thickness, obtained following the local correction of the movement of the measuring device 10. This minimum squared thickness must be less than a predefined threshold to determine that a potential area of interest is indeed an area of interest, i.e. for example a planar area.
[0062] Advantageously, the predefined threshold represents an expected squared thickness, which takes into account measurement noise but not reconstruction errors. In particular, LiDAR manufacturers indicate a distance uncertainty σr, an azimuth uncertainty σ4, and an elevation uncertainty Résose as a standard deviation or maximum error value. From this, a global squared measurement error equal to σ² + σ²(rstei) can be deduced, with r the measured distance and the elevation.
[0063] The overall squared measurement error provides a possible threshold value if the uncertainties , 89, Se given by the manufacturer represent maximum errors. If the uncertainty values 8f, 80, Se given by the manufacturer are standard deviations, it is possible to multiply them by three to obtain a threshold representative of the expected squared thickness.
[0064] According to another embodiment, in which the thickness, and not the thickness squared, is compared to a predefined threshold, the square root of criterion C' is to be compared with the square root of the overall root mean square measurement error. When the uncertainty values gf, 89, Se given by the manufacturer are standard deviations, the uncertainty values are also to be multiplied by three to obtain a threshold representative of the expected thickness.
[0065] In the following description, and in a non-limiting manner, the criterion C' corresponds to or is representative of a squared thickness and the predefined threshold is representative of the overall squared measurement error.
[0066] Thus, in order to determine whether the potential area of interest is indeed an area of interest, an optimization problem can be solved. It takes the form:
[0067] C - min C(x1(Amouv), .x?(Amouv), ...)(1)
[0068] Where Amouv is a variation of the motion of the measuring device 10 at each point x, and where xi=i>2,...(Amouv) is the position of point i after application of a variation of motion Amouv. The initial position of the point x;(0), for Amouv=0, is simply denoted x;. In other words, the processor 11 searches for a correction of all positions equivalent to a variation of motion Amouv of the measuring device 10. The minimum value of 0(^(Amouv), %?(Amouv), ... ) is denoted C'.
[0069] Solving the optimization problem allows us to obtain the value of criterion C' and to determine if the potential area of interest is indeed an area of interest by comparing C' to the predefined threshold.
[0070] According to a second embodiment in which only the local speed is corrected, the optimization problem takes the form:
[0071] C = minC(x,(Av), x9(Av), ...) Av
[0072] Where Av is a variation in the speed of the measuring device 10 at each point xi5 and where Xj(Av) can be written as Xj-fAv, with i=1, 2, .... In other words, the processor 11 searches for a correction of all positions equivalent to a speed variation Av of the measuring device 10. The minimum value of C(Xj(Av), X2(Av), ... ) is denoted C' ( Xj, X2, .. •, . ) •
[0073] Moreover, in an embodiment where only the local velocity is corrected and in which criterion C corresponds to the smallest eigenvalue of the covariance matrix 1 y ( y - - X ) ( Y - Y ) ^' result of the optimization problem (1) can then be obtained directly: 100741 C'(x
[0075] With x> the three-coordinate points of the initial point set representing a potential area of interest, N the number of points in the initial point set considered, x is the average position of said points defined by x — A is the average position of said instants defined by / — the sum 1 y ( x \ ( x _ \ r is the covariance matrix of the points, _ E;( trt ) û and the ^Y The function A( ■ ) returns the smallest eigenvalue of the matrix given as an argument.
[0076] According to this implementation, once the value C' has been determined, it is then possible to determine whether the potential area of interest is indeed an area of interest. In this example, the area of interest considered is a planar area. Indeed, the covariance matrix characterizes the distribution of points in the local orthogonal coordinate system locating the points of the initial set of points representing the potential area of interest. Thus, if C' is less than the predefined threshold, it is determined that the area is planar.
[0077] If the potential area of interest is not actually one, that is to say if the potential area of interest is not planar, C' will take a value greater than the predefined threshold and this value will increase with the actual thickness of the point cloud formed by the set of points associated with the potential area of interest considered.
[0078] Thus, according to the second embodiment, the calculation of the value of criterion C' is direct and does not require an optimization process. This advantage is specific to the case where a speed correction is sought. This embodiment also allows to greatly reduce calculation times as well as the memory allocated to calculations.
[0079] Moreover, since the predefined threshold represents an expected quadratic thickness affected by measurement noise but not by reconstruction errors, it is determined with precision whether a potential area of interest is indeed an area of interest.
[0080] The search for areas of interest representing flat areas in the environment was described above. As mentioned above, these areas may include, in particular, a wall, a corner of a wall, a roadside, a sidewalk, but also a flat traffic sign.
[0081] Generally, such traffic signs are mounted on cylindrical posts. It may then be useful to define the sign and post assembly, or simply the post, as an area of interest.
[0082] Regarding planar areas, one criterion for indicating that an area is planar is to determine that all points of that area lie in the same plane, as described above. It is possible to define such a criterion for more complex shapes such as cylinders.
[0083] The question then becomes whether a cylinder can pass through all the points of the initial point set corresponding to the cylindrical element. Such an element could be a post, but also a tree trunk, a stake, etc.
[0084] In other words, a cylinder is defined by its axis u, its center c and its radius r. It can be determined that a cylinder passes through all the points of an initial set of points, or passes at least close to a majority of the points if the distance between each of these points, or the vast majority of the points, and this cylinder is small, or even equal to zero.
[0085] The distance from a point x to such a cylinder is written l / 77 . <2 h \ > >2 i. It is possible to determine that a cylinder passes through or near all, or most, of the points if the average of the squared distances is less than a predefined threshold corresponding to the root mean square error of the Lidar distance measurements, generally provided by the Lidar manufacturer. This error is usually on the order of 1 cm but can vary depending on the Lidar used.
[0086] According to this embodiment, to determine if the parameters of the cylinder u, r, c correspond to the points xb ... xn, the root mean square distance from the points xb ... xn to the cylinder, representative of a thickness, is calculated according to:
[0087] iv I fi----iL.....ÏT77----"TT i2 NLjVlurHi - Hw7Urc)ll -r|
[0088] However, only the points x1 of the potential area of interest are known. The parameters u, c, and r of the cylinder that the points x1 could form are not known. The One way to determine if a cylinder passes through all the points, or at least the majority of the points, is to say that there exists at least one cylinder for which C will be minimal, or even equal to zero, where C is written: [oo89] x irir----.............7----r C(%1? ....%) read -r|
[0090] Finally, the minimum value of C when applying the motion correction AMouv, denoted C', is the criterion for determining whether the potential area of interest is indeed an area of interest. C' is obtained according to:
[0091] (7(¾ = .---------------------------------------------------------— 2 mi n min^ E • JH x,( Amouv) - d I -H uT ( x(Amouv) - c ) 1I - r
[0092] Where Amouv is a correction of the movement of the measuring device along the potential area of interest, i.e. of its trajectory and / or its speed.
[0093] If the result of the double minimization is less than a predefined threshold such as the mean squared error of the Lidar distance measurements, then the potential area of interest is determined to be effectively an area of interest, here a cylindrical shape.
[0094] If the result of the double minimization is greater than a predefined threshold such as a multiple of the mean squared error of the Lidar distance measurements, it is then determined that the potential area of interest represented by the initial set of points considered is not an area of interest.
[0095] It should be noted that, in cases where a correction of the local velocity of the measuring device has been described, a correction of the local trajectory of the measuring device can be made as an alternative and / or in addition.
[0096] Finally, according to a last embodiment, the criterion for determining whether a potential area of interest is indeed an area of interest is the intensity distribution of the points in an initial set of points. Here, the intensity represents the fraction of the laser light energy emitted by the lidar 101 and reflected by the material struck.
[0097] Such an area of interest may be an inscription, for example an inscription on a wall, a vehicle, etc., or an image represented on a panel, for example an advertising panel or a traffic sign.
[0098] In this embodiment, it is a local trajectory of the measuring device 10 along the potential area of interest that is corrected for each initial set of points representing a potential area of interest. The task then becomes to determine whether, for a potential area of interest, the variations in the arrangement of the points carrying an intensity are correlated with variations in the position of the measuring device. In other words, the aim is to determine whether a local variation in the trajectory of the measuring device along the potential area of interest, which includes points carrying different intensities, results in a different distribution of points, thus defining a potential area of interest as indeed an area of interest. Specifically, such a difference in the distribution of intensity points can make an inscription legible that would otherwise be illegible without local correction of the measuring device along that area.
[0099] Such a criterion E can be written:
[0100] he 2
[0101] Where li is the intensity measured at a point L of the initial set of points, J — is the average intensity in the area and || ■ || is the classical Euclidean norm of a vector, here of size 3.
[0102] It is determined that a potential area of interest is indeed an area of interest when the value of criterion E reaches, for example, 0.5.
[0103] More specifically, for all embodiments described with reference to step S3, the calculation of the value of the criterion involves making corrections to the local movement of the measuring device 10, here on the speed or trajectory of the measuring device 10. The calculation of the criterion makes it possible to determine with certainty whether a zone of interest can actually be obtained by locally modifying the movement of the measuring device 10 along the potential zone of interest.
[0104] At the end of step S3, all areas of interest have been identified.
[0105] Then, an optional step S4 can be implemented, in which, following the determination that a potential area of interest is indeed an area of interest, it is determined whether all the points in the initial point set representing the area of interest actually belong to the area of interest. Indeed, in some cases, points far from the area of interest may have been included in the area of interest by mistake. The task then is to find these points and remove them.
[0106] To do this, it is possible to calculate the proximity of each point in the initial set of points to the area of interest, and then to remove the points that are too far from this area of interest.
[0107] More specifically, considering areas of interest such as wall corners, signs and / or signposts and flat areas.
[0108] In particular, for road signs, flat areas, and wall corners, it is necessary to determine whether all the points of the initial point set are indeed part of the same plane determined in step S3. Points not belonging to the plane elements formed by the area of interest are removed.
[0109] Similarly, for more complex areas of interest such as signposts, it is necessary to determine whether all the points in the initial point set are part of the cylinder determined in step S3. Points not part of the cylinder formed by the area of interest are removed.
[0110] In step S5, a joint correction is applied to the initial point sets effectively representing an area of interest by applying a global correction to a movement of the measuring device 10 in the environment. More precisely, the global correction is applied to a trajectory of the measuring device 10 in the environment.
[0111] This global, rather than local, correction of the trajectory of the measuring device 10 is due to the fact that it is a matter of finding a single trajectory that allows us to obtain a map of the environment on which all the identified areas of interest are represented, and not of finding different trajectories for each area of interest.
[0112] This is solved by means of an optimization problem. The trajectory is obtained by defining an optimization problem with two components: the structural criterion of the regions of interest (flatness, for example) and the correspondence between the chosen trajectory and the additional available measurements, obtained, for example, from an inertial measurement unit. Solving such an optimization problem is well-established, and off-the-shelf solvers offer this solution in a fairly generic framework, for example, the GTSAM solver mentioned in the following publication:
[0113] Kaess, M., Johannsson, H., Roberts, R., lia, V., Leonard, JJ, & Dellaert, F. (2012). iSAM2: Incremental smoothing and mapping using the Bayes tree. The International Journal of Robotics Research, 31(2), 216-235.
[0114] Then, once the overall trajectory of the measuring device has been corrected, the environment map is constructed with the corrected initial point sets that effectively represent an area of interest and with the initial point sets that do not represent an area of interest.
[0115] In one embodiment, the mapping of the environment and the correction of the global trajectory of the measuring device 10 are carried out simultaneously.
[0116] Figure 3 illustrates another embodiment according to the invention. Steps S1 to S5 described in relation to Figure 2 are identical. According to this embodiment, steps S2 to S4 are iterated until a predefined criterion is met in step S6.
[0117] This predefined criterion can be reached when, between two iterations, the correction of the movement of the measuring device no longer changes.
[0118] This arrangement allows for a more precise correction of the local trajectory, and / or for a more certain determination of whether an area of interest is indeed an area of interest or not.
Claims
Claims
1. System (1) for constructing a map of an environment, the system (1) comprising: - a measuring device (10) comprising at least one lidar (101) configured to acquire a point cloud representative of said environment in which the measuring device (10) moves, - a processor (11) configured to * preselect a plurality of initial point sets in the point cloud, each initial point set of the plurality of initial point sets being representative of a potential area of interest of the environment, * for each initial point set, search for a local correction to be made to a movement of the measuring device along each potential area of interest represented by the initial point set considered in order to determine whether each initial point set actually represents an area of interest,*jointly correct the initial point sets actually representing an area of interest, by applying a global correction to a movement of the measuring device in the environment, Construct the environment map with the corrected initial point sets actually representing an area of interest and the initial point sets not representing an area of interest.,
2. The system (1) of claim 1, wherein the processor (11) is configured to iterate the steps of preselection, local correction and joint correction until a predefined criterion is met.
3. System (1) according to claim 1 or 2, in which the processor (11) is configured to carry out the local correction searches to be made to each initial set of points in parallel.
4. System (1) according to one of claims 1 to 3, in which the local correction to be made to a movement of the measuring device (10) comprises the correction of a local trajectory of the measuring device (10) along each potential area of interest and in which the joint correction comprises the correction of a global trajectory of the measuring device in the environment.
5. System (1) according to one of claims 1 to 4, in which the local correction to be made to a movement of the measuring device (10) comprises the correction of a local speed of the measuring device (10) along each potential area of interest and in which the joint correction comprises the correction of an overall trajectory of the measuring device in the environment.
6. System (1) according to one of claims 1 to 5, in which the global correction to be made to a movement of the measuring device (10) in the environment is determined by solving an optimization problem.
7. System (1) according to claim 5, wherein the areas of interest comprise a planar area such as a road, a traffic sign or a wall, the processor (11) being configured, for each potential area of interest corresponding to a potential planar area, to calculate a minimum thickness C' of said potential planar area obtained after the local correction of the movement of the measuring device, said minimum thickness being obtained by direct resolution of: C(X, T) where N is the number of points xi in the initial set of points considered, x — - xt is the average of the points xt of the initial set of points, l — ~^.ti is the average of the instants 1; associated with said points of the initial set of points, is intermediate of calculation, s~ ---- > V is the covariance matrix, and the function A( ■ ) returns the smallest eigenvalue of the matrix given as argument, and where the processor (11) is configured to determine that the potential flat area is indeed an area of interest if C' is less than a predefined threshold.
8. System (1) according to one of the preceding claims, in which the areas of interest comprise a cylindrical area such as a post, the processor (11) being configured, for each potential area of interest corresponding to a potential cylindrical area, to determine whether there is a correction of a local movement (Amouv) of the measuring device for which the distance is minimal between each of the N points xi of the initial set of points considered and a cylinder of center c, axis u and radius r according to a double minimization: ■________________________________________________________________________________ 2 min min4J H x.( Amouv) - c IT -11 uT ( x / Amouv) - c ) IT - r I where Amouv is a variation of the movement of the measuring device 10 at each point xi5 and where xi=i, 2, ...(Amouv) is the position of the point i after applying a variation of movement Amouv, the processor being further configured to determine that a potential cylindrical area of interest is indeed a cylindrical area of interest if a result of the double minimization is less than a predefined threshold.
9. System (1) according to one of the preceding claims, in which for each initial set of points effectively representing an area of interest, the processor (11) is configured to delete the points present in the initial set of points but not belonging to said area of interest.
10. A method of constructing a map of an environment, the method comprising: - The acquisition, by a measuring device (10) comprising at least one lidar (101), of a point cloud representative of said environment in which the measuring device moves, - The implementation, by a processor (1): *A preselection of a plurality of initial point sets in the point cloud, each initial point set of the plurality of initial point sets being representative of a potential area of interest in the environment, *for each initial set of points, a search for a local correction to be made to a movement of the measuring device (10) along each potential area of interest represented by the initial set of points considered in order to determine whether each initial set of points actually represents an area of interest, *a joint correction of the sets of initial points effectively representing an area of interest, by applying a global correction to a movement of the measuring device (10) in the environment, *a construction of the map of the environment with the corrected sets of initial points effectively representing an area of interest and the sets of initial points not representing an area of interest.
