Method for estimating the dimensions of an object from points acquired by a lidar sensor

The method corrects lidar sensor dimension estimates using intensity errors and cost function minimization, addressing resolution limitations to enhance safety in autonomous driving by accurately determining object dimensions.

FR3149390A1Active Publication Date: 2024-12-06CONTINENTAL AUTONOMOUS MOBILITY GERMANY GMBH
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
FR2023005420
Authority / Receiving Office
FR · FR
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-31
Publication Date
2024-12-06
Estimated Expiration
2043-05-31

AI Technical Summary

Technical Problem

The determination of the dimensions of objects, particularly small objects located several tens of meters from a vehicle, is limited by the resolution of a lidar sensor, leading to underestimation or overestimation of these dimensions.

Method used

A method for estimating the dimensions of an object using a lidar sensor that involves acquiring a set of points with three-dimensional coordinates and intensity values, selecting a group of presumed points belonging to the same object, estimating dimensions from these points, and correcting the estimates using intensity errors through a cost function minimization process.

Benefits of technology

The method provides precise estimation of object dimensions, reducing the likelihood of collisions by accurately determining vehicle trajectories based on corrected dimensions, enhancing safety in autonomous driving scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000000_0000_ABST
    Figure 00000000_0000_ABST
Patent Text Reader

Abstract

Examples present a process (100) of at least one dimension belonging to a reflective surface of an object comprising: an acquisition (110) by a lidar of a set of points; a selection (120) of light beams from a group of points of the same object; an estimation (130), from the three-dimensional coordinates of the points, of the values ​​of a first and a second dimension of the object; an estimation (140), from the intensity values ​​of the points, of a reflectivity value of the object; then for at least one specific point of the group: an estimation (150) of a cross-section of a light beam; an estimation (160) of a theoretical intensity value received by the lidar sensor following the emission of the beam; a determination (170) of an intensity error associated with the specific point from a difference between the estimated theoretical intensity and the measured intensity;and a correction (180) of the estimation of the value of the first dimension from the intensity error. Figure of the abstract: [Fig. 3];
Need to check novelty before this filing date? Find Prior Art

Description

Title of the invention: Method for estimating the dimensions of an object from points acquired by a lidar sensor Technical field

[0001] The present disclosure relates to the field of processing data acquired by a lidar sensor. Prior art

[0002] A lidar (Light Detection And Ranging) sensor is a sensor emitting light waves and determining, from the reflection of these light waves, a matrix of points representing a lidar environment.

[0003] A lidar sensor can be used to perform measurements of an object in its environment by acquiring a point cloud representing the environment of the lidar sensor, the object then being represented by a plurality of points in the point cloud. In particular, when the lidar sensor is on board a vehicle, it can for example be used to detect objects present on the road. In the context of driving assistance functions, in particular autonomous driving functions, detecting and determining the dimensions of an object present on the road allows the vehicle to react accordingly. In particular, when it is a small object, the precise dimensions of this object can make it possible to determine whether or not the vehicle can pass over the object without risking a collision.

[0004] However, the determination of the dimensions of objects, and more particularly of small objects located several tens of meters from the vehicle, is limited by the resolution of the lidar sensor, so that the objects can sometimes be under or oversized.

[0005] The present disclosure improves this situation. Summary

[0006] In this regard, a method is proposed for estimating at least one dimension belonging to a reflection surface of an object, the method being implemented by a device comprising at least one lidar sensor, a computer and a memory, the memory storing a model of the object comprising a first dimension and a second dimension of the reflection surface of the object and a reflectivity of the reflection surface of the object; the method comprising: - an acquisition by the lidar sensor of a set of points, each point being associated with three-dimensional coordinates in space and an intensity value; - a selection from the set of points of a group of points presumed to be ap to go to the same object; - an estimate, from the three-dimensional coordinates of the points of the group of points, of the values ​​of a first and a second dimension of the object corresponding respectively to the first and second dimensions of the reflection surface of the model of the object; - an estimate, from the intensity values ​​of the group of points presumed to belong to the object, of a reflectivity value of the object corresponding to the reflectivity of the reflection surface of the model of the object; then for at least one specific point of the group of points presumed to belong to the object: • an estimate of a section of a specific light beam whose reflection by the reflection surface of the object made it possible to obtain the specific point, from the estimated dimensions of the reflection surface of the object and a divergence of the light beam determined from a distance between the specific point and the lidar sensor; • an estimate of a theoretical intensity value received by the lidar sensor following the emission of the specific light beam, from the estimated specific light beam section and the estimated reflectivity value of the reflection surface; • a determination of an intensity error associated with the specific point from a difference between the estimated theoretical intensity value and the measured intensity value associated with the specific point following the emission of the specific light beam; and - a correction of the estimate of the value of the first dimension from the intensity error associated with at least one specific point.

[0007] Optionally, a correction of the estimation of the value of the first dimension from the intensity error associated with at least one specific point comprises a minimization of an error of a cost function calculated from the intensity error associated with at least one specific point by iteratively varying the value of the first dimension.

[0008] Optionally, the method further comprises a correction of the estimation of the value of the second dimension from the intensity error associated with at least one specific point, the correction comprising a minimization of an error of a cost function calculated from the intensity error associated with at least one specific point by iteratively varying the value of the second dimension.

[0009] Optionally, the method further comprises a correction of the estimation of the reflectivity value from the intensity error associated with at least one specific point, the correction comprising a minimization of an error of a cost function calculated from the intensity error associated with at least one specific point in iteratively varying the reflectivity value.

[0010] Optionally, the cost function corresponds to the sum of the intensity errors of the specific points for which the intensity error was determined.

[0011] Optionally, the first dimension of the object model corresponds to a height of the reflection surface of the object and wherein the second dimension of the object model corresponds to a width of the reflection surface of the object.

[0012] Optionally, the theoretical intensity value received by the lidar sensor following the emission of the specific light beam is estimated from a function characterizing the intensity received by the lidar following the reflection of a light beam emitted by the lidar sensor on a surface as a function of: a. a distance between this surface and the lidar sensor, b. of a reflectivity of this surface, and c. of a section of the light beam reflected on this surface.

[0013] Optionally, the plurality of points is enriched by new acquisitions from the lidar sensor and the method is implemented iteratively from the enriched plurality of points.

[0014] Optionally, the method further comprises determining a trajectory of a vehicle carrying the device from the corrected estimate of the value of the first dimension.

[0015] The application also relates to a device comprising at least one lidar sensor, a computer and a memory, the device being configured to implement any of the methods presented by the present disclosure. In particular, the device can be on board a vehicle.

[0016] The application further relates to a computer program product comprising instructions for implementing any of the methods presented by the present disclosure when this program is executed by a processor.

[0017] Finally, the application relates to a non-transitory recording medium readable by a computer on which is recorded a program for implementing any of the methods presented by the present disclosure when this program is executed by a processor.

[0018] The method according to the present disclosure thus makes it possible to estimate the length of at least one side of the reflection surface of an object captured by the lidar sensor precisely and in particular more precisely than a conventional estimation based solely on the three-dimensional coordinates of the points presumed to belong to the object, these dimensions possibly being under- or over-evaluated due to the resolution of the lidar sensor. In certain options, the method comes to estimate the length of the two sides of the reflection surface of the object (width L and height h). These estimations of the dimensions of the reflection surface of the object can in particular be used to determine a vehicle trajectory adapted to the precise dimensions of the object, which can help avoid situations in which a vehicle trajectory determined on the basis of underestimated dimensions of the object could be detrimental to the safety of the vehicle and road users. Brief description of the drawings

[0019] Other characteristics, details and advantages will appear on reading the detailed description below, and on analyzing the attached drawings, in which: [Fig.l] [Fig.l] schematically represents a device allowing the realization of a process for estimating a dimension belonging to a reflection surface of an object. [Fig.2] [Fig.2] represents an example of a lidar sensor on board a vehicle. [Fig.3] [Fig.3] represents an example of a method for estimating at least one dimension belonging to a reflection surface of an object. [Fig.4] [Fig.4] represents an example of a section of light beams emitted by the lidar sensor and reflected by the reflective surface of an object. Description of the embodiments

[0020] An example of a device 1 is now described with reference to [Fig.l] for carrying out a method for estimating a dimension belonging to a reflection surface of an object.

[0021] The device 1 can be adapted to be mounted on a vehicle 2.

[0022] The device 1 comprises a lidar sensor 10 (Light Detection And Ranging in English), a calculator 11 and a memory 12.

[0023] The memory 12 can store the code instructions executed by the computer 11 and making it possible to control the acquisition of data by the lidar sensor 10 as well as their processing. The computer therefore has access to the information stored in memory. The memory 12 can also be adapted to store the data acquired by the lidar sensor 10.

[0024] The memory 12 may for example comprise a ROM (Read-Only Memory), a RAM (Random Access Memory), an EEPROM (Electrically Erasable Programmable Read-Only Memory) or any other type of suitable storage means. The memory may for example comprise optical, electronic or magnetic storage means.

[0025] The lidar sensor 10 is adapted to acquire three-dimensional point clouds (x, y, z) representing the environment of the vehicle 2 and in particular representing the environment in front of the vehicle. The lidar sensor 10 can thus be oriented to emit light beams representing the environment in front of the vehicle, that is to say in its direction of movement represented by the arrow d in [Fig.2]. The lidar sensor 10 can for example be mounted on the body at the front of the vehicle 2.

[0026] With respect to three-dimensional coordinates, the x coordinate of a point corresponds to an abscissa coordinate of the point relative to the sensor 10. The y coordinate of a point corresponds to an ordinate coordinate of the point relative to the sensor 10. The z coordinate of a point corresponds to an elevation coordinate of the point relative to the lidar sensor 10. In [Fig. 2], and assuming the contours of the vehicle 2 and the lidar sensor 10 are shown in top view, the abscissa axes X and ordinate axes Y of the lidar sensor 10 are shown. The elevation axis Z of the sensor 10 is perpendicular to the X and Y axes. In this [Fig. 2], the lidar sensor 10 is therefore mounted at the front of the vehicle and is oriented to emit light beams in the direction of travel of the vehicle.

[0027] Each point acquired by the lidar sensor 10 is also associated with an intensity value. This is the intensity received by the lidar sensor after reflection of the light beam on a surface.

[0028] With reference to [Fig. 3] an example of a method 100 is now presented that the device 1 according to the present disclosure is configured to implement. In particular, the device 1 is configured to implement a method for estimating at least one dimension belonging to a reflection surface of an object. The reflection surface of the object corresponds to a surface of the object having reflected the light beams emitted by the lidar sensor 10. A dimension of this surface here designates a height or a width of this surface. It is therefore a question of estimating the length of at least one side of this surface precisely and in particular more precisely than a conventional estimation based solely on the three-dimensional coordinates of the points presumed to belong to the object.Indeed, such an estimation is notably limited by the resolution of the lidar sensor 10 so that for objects located at a long distance from the lidar sensor, the error rate in the estimated dimensioning of the object relative to the size of the object can be significant.

[0029] To enable this estimation, the memory 12 stores a model of the object comprising a first dimension, for example a height h, and a second dimension, for example a width L, of the reflection surface of the object. The model of the object also comprises a reflectivity of the reflection surface of the object. The model of the object therefore comprises three parameters, the first and second dimensions of the reflection surface of the object, and the reflectivity of the reflection surface of the object. The reflectivity is data reflecting the more or less reflective nature of the reflection surface of the object to light and therefore the more or less reflective nature of the reflection surface of the object to the light beams emitted by the lidar sensor 10. One of the objectives of the method is to determine precise values ​​of the reflection surface of the object by estimating the values ​​of the model parameters as developed in the rest of the document.

[0030] In this case, the method 100 relies on the fact that a light beam emitted by a lidar sensor 10 diverges during its propagation. It is therefore not concentrated at a single point at the moment of striking the surface on which it is reflected (referred to as the reflection surface hereinafter). It is in reality a section of the light beam which strikes the reflection surface. This section generally has an elliptical shape when the entire beam is reflected by the reflection surface. Furthermore, this section, insofar as the entire beam is reflected by the reflection surface, is all the larger the further the reflection surface is from the lidar sensor 10 since the light beam diverges more and more as it moves away from the lidar sensor 10.Therefore, the intensity of the light beam reflected by the reflection surface of the object and captured by the lidar sensor 10 depends on the section of the light beam actually reflected by the reflection surface of the object. In particular, the reflection section of a light beam emitted for a point acquired on an edge of the reflection surface of an object at a given distance will not correspond to a complete theoretical section of the light beam hitting this surface, which means concretely that a complementary section of the light beam will not be reflected by the reflection surface. This situation is notably represented in [Fig.4] illustrating a reflection surface Sr and five light beams Fl to F5, three of which are completely reflected by the reflection surface Sr (namely the light beams F2, F3 and F4) and two of which are partly reflected by this surface Sr (namely the light beams Fl and F5).Here, the intensity measured by the lidar sensor 10 will be lower for the points F1 and F5 than for the points F2 to F4 if it is assumed that the entire reflection surface Sr has the same reflectivity. As explained below, the method 100 cleverly uses the intensity information of the points associated with the reflection surface Sr of an object to correct a first estimate of at least one dimension of this reflection surface Sr made from the three-dimensional coordinates of the points. In this way, the dimensions of the reflection surface Sr of an object captured by a lidar sensor 10 are no longer only estimated from the three-dimensional coordinates of the points belonging to this object but also from the intensity associated with these points by using the correlation between this intensity and the reflection section of the light beams associated with these points on the reflection surface Sr.Consequently, the method 100 makes it possible to reduce the number of collisions due to objects whose dimension estimation accuracy is limited due to the resolution of the lidar sensor 10. In particular, in the context of autonomous driving functions, the vehicle may be required to modify its tra . jectory when an object, for example a tire, is present on the road in front of it. In this case, the precise determination of the width and height of the object is essential to determine the trajectory of the vehicle, whether it is to go around the object or simply to pass over the object, between the wheels of the vehicle when the height of the object allows it. This is particularly advantageous when the vehicle is traveling at high speed and a decision must be made extremely quickly. It is understood that a decision such as emergency braking or sudden swerving of the vehicle that would not be taken to the extent that the dimensions of the object detected by the lidar sensor would have been underestimated raises safety questions to the extent that a more precise estimation, such as proposed by the method according to the present disclosure, would have made it possible to make the appropriate decision.

[0031] The method 100 thus comprises an operation 110, as illustrated by [Fig. 3], of acquisition by the lidar sensor 10 of a set of points. Each point acquired by the lidar sensor 10 is associated with three-dimensional coordinates (x, y, z) in space and with an intensity value.

[0032] The method 100 comprises an operation 120, as illustrated by [Fig. 3], of selecting, from the set of points acquired by the lidar sensor 10, a group of points presumed to belong to the same object.

[0033] In examples, the selection 120 of the group of points is performed based on the nearest neighbor criterion. In other words, the selection 120 of the group of points may comprise a selection of a set of points whose distance with their nearest neighbor is less than a predetermined distance threshold.

[0034] Of course, the method 100 may comprise the selection of several groups of points if several groups of points can be determined as presumed to each belong to the same object. The following operations of the method 100 may then be implemented for each of the groups of points determined at the end of the operation 120.

[0035] The method 100 comprises an operation 130, as illustrated by [Fig. 3], of estimating the values ​​of a first dimension and a second dimension of the object. The first dimension and the second dimension of the object correspond respectively to the first dimension and the second dimension of the reflection surface of the model of the object. These values ​​are estimated from the three-dimensional coordinates of the points of the group of points selected at the end of the operation 120.

[0036] In examples, the values ​​of the first and second dimensions may be estimated from the points of the group of points having the most extreme coordinates on the abscissa axis (x) and on the elevation axis (z). In particular, in examples, the value of the first dimension, when it is the height h of the reflection surface of the object Sr, may correspond to a difference between the largest elevation value associated with a point of the group of points and the lowest elevation value associated with a point of the group of points. In examples, the value of the first dimension, when it is the width of the reflection surface of the object Sr, may correspond to a difference between the largest abscissa value associated with a point of the group of points and the lowest elevation value associated with a point of the group of points. Thus, the operation 130 makes it possible to obtain a first estimate of the dimensions of the reflection surface of the object detected by the lidar sensor 10 from the three-dimensional coordinates of the points of the group of points presumed to belong to the object.

[0037] The method 100 comprises an operation 140, as illustrated by [Fig. 3], of estimating a reflectivity value of the object. The reflectivity value of the object corresponds to the reflectivity of the reflection surface of the model of the object. This reflectivity value of the object is estimated from the intensities of the points of the group of points presumed to belong to the object. In examples, a reflectivity of a point of the set of points can be determined from a distance between the point and the lidar sensor, an intensity of the light beam used to obtain the point and an intensity received by the lidar sensor following the reflection of this beam. Thus, in examples, the reflectivity of the object can be determined as corresponding to an average of the reflectivities of the points of the group of points presumed to belong to the object.In other examples, the reflectivity of the object may be determined as corresponding to an average of the reflectivity values ​​of a selection of points belonging to the group of points. In these other examples, it is a matter of selecting a reduced number of points to determine the reflectivity of the object, for example by removing the points whose intensity values ​​are clearly erroneous. It is therefore assumed in this operation that the reflecting surface of the object has a substantially homogeneous reflectivity.

[0038] Following operation 140, several operations will be performed for at least one specific point of the group of points. These operations are advantageously performed for each point of the group of points presumed to belong to the object.

[0039] The method 100 comprises an operation 150, as illustrated by [Fig. 3], of estimating a section of a specific light beam whose reflection by the reflection surface of the object has made it possible to obtain the specific point. This operation therefore involves estimating the section of the light beam having actually been reflected by the light beam. As explained previously, it is understood that a section associated with a point acquired by the lidar sensor 10 and corresponding to a point positioned in the middle of the reflection surface Sr of the object will correspond to the entire section of the light beam emitted by the lidar sensor 10. Conversely, a section associated with another point acquired by the lidar sensor 10 and corresponding to a point positioned on one end of the reflection surface Sr of the object will correspond to a section amputated with respect to the entire section of the light beam emitted by the lidar sensor 10. These are the situations explained previously with reference to [Fig.4],

[0040] In examples, the section of the specific light beam whose reflection by the reflection surface of the object made it possible to obtain the specific point is estimated from the estimated dimensions of the reflection surface of the object and a divergence of the light beam determined from a distance between the specific point and the lidar sensor. Indeed, the divergence of the light beam emitted by the lidar sensor 10 increases with the distance. Consequently, by knowing the orientation of the lidar sensor and the distance at which the specific point is located relative to the lidar sensor 10 using its three-dimensional coordinates, it is possible to determine the theoretical section of the light beam at this distance.It is therefore possible to determine the projection of this section onto the reflection surface Sr of the object by knowing the orientation of the specific light beam and the dimensions of this reflection surface Sr, which were estimated during operation 130. A distance between the specific point and the lidar sensor 10 corresponds to the Euclidean distance between the specific point and the lidar sensor 10.

[0041] The method 100 comprises an operation 160, as illustrated by [Fig. 3], of estimating a theoretical intensity value received by the lidar sensor following the emission of the specific light beam, from the estimated section of the specific light beam and the estimated reflectivity value of the reflection surface. This step involves estimating a theoretical intensity that the lidar sensor 10 should have received after reflection of the light beam on the reflection surface Sr of the object with regard to the section of the light beam reflected on this surface estimated during the operation 150 and with regard to the reflectivity of this surface estimated during the operation 140.

[0042] Thus, when the intensity estimated for the specific point does not correspond to the intensity actually measured by the lidar sensor 10 for the specific point, the section of the light beam estimated as having been reflected by the reflection surface includes an error since it is from this section that the theoretical intensity is calculated. However, the section of the light beam is itself estimated from the dimensions of the reflection surface so that it is these estimated dimensions of the reflection surface that are erroneous. Thus, by comparing the theoretical intensity value with the intensity value actually measured by the lidar sensor 10 for the specific point, the method makes it possible to obtain information on the accuracy of the dimensions of the reflection surface estimated during operation 130.

[0043] In examples, the theoretical intensity value received by the lidar sensor following the emission of the specific light beam is estimated from a function characterizing the intensity received by the lidar sensor following the reflection of a light beam emitted by the lidar sensor on a surface as a function of: a. a distance between this surface and the lidar sensor, b. surface reflectivity, and c. of a section of the light beam reflected by this surface.

[0044] This function can for example be predetermined empirically from acquisitions of one or more lidar sensors, advantageously from acquisition of the lidar sensor 10, by varying the reflectivity of the reflection surfaces of the objects, their position relative to the lidar sensor and the section of the light beams reflected by the reflection surface of these objects.

[0045] Thus, by initializing the different parameters of this function with the distance between the specific point and the lidar sensor, the reflectivity of the reflection surface of the object estimated during operation 140 and the section of the light beam reflected by the reflection surface of the object having made it possible to obtain the specific point estimated during operation 150, the function makes it possible to calculate the theoretical intensity value of operation 160.

[0046] The method 100 comprises an operation 170, as illustrated by [Fig. 3], of determining an intensity error associated with the specific point. The intensity error associated with the specific point is obtained from a difference between the theoretical intensity value estimated during the operation 160 and the intensity value actually measured by the lidar sensor 10 and associated with the specific point following the emission of the specific light beam. As explained previously, this intensity error reflects an error in the estimation of the dimensions of the reflection surface Sr of the object.

[0047] Operations 140, 150, 160 and 170 are therefore carried out for at least one specific point and advantageously for each point of the group of points presumed to belong to the same object. Consequently, at the end of operation 170, at least one intensity error associated with a specific point has been calculated.

[0048] The method then comprises an operation 180, as illustrated in [Fig.3], of correcting the estimation of the value of the first dimension from the intensity error associated with at least one specific point.

[0049] In examples, the operation 180 of correcting the estimate of the value of the first dimension comprises correcting the estimate of the value of the first dimension in order to reduce the at least one intensity error.

[0050] In first examples, the operation 180 of correcting the estimation of the value of the first dimension from the intensity error associated with at least one specific point may include minimizing an error of a cost function calculated from the intensity error associated with at least one specific point by iteratively varying the value of the first dimension. In these first examples, the cost function expresses the error to be minimized as a function of the value of the first dimension of the object model.

[0051] In second examples, complementary to the first examples, the method may comprise an operation 181 of correcting the estimation of the value of the second dimension from the minimization of the error of the cost function calculated from the intensity error associated with at least one specific point by iteratively varying the value of the second dimension. In these second examples, the cost function expresses the error to be minimized as a function of the value of the first dimension and the value of the second dimension of the model of the object.

[0052] In third examples, complementary to the first examples and possibly the second examples, the method may comprise an operation 182 of correcting the estimation of the value of the reflectivity from the minimization of the error of the cost function calculated from the intensity error associated with at least one specific point by iteratively varying the value of the reflectivity. In these third examples, the cost function expresses the error to be minimized as a function of the value of the first dimension and the value of the reflectivity of the model of the object. When these third examples are implemented in complementarity with the second examples, the cost function expresses the error to be minimized as a function of the value of the first dimension, the value of the second dimension and the value of the reflectivity of the model of the object.

[0053] It is to be understood by iterative variation of a parameter of the model of the object the fact of choosing a value of this object at a given iteration and of implementing again the operations 140 to 170 while taking into consideration this new value of the parameter of the model.

[0054] In the first, possibly second and / or third examples, the error to be minimized of the cost function may correspond to the sum of the intensity errors associated with the specific points for which the intensity error was determined during operation 170.

[0055] Thus, by iteratively varying the value of the first dimension, by optionally iteratively varying the value of the second dimension and / or the value of the reflectivity of the model of the object, and by implementing operations 140 to 170 again at each iteration, it is possible, with an optimization algorithm, to minimize the error of the cost function, which is determined from the at least one intensity error determined at the end of operation 170. Minimize the error of the cost function by an iterative variation of the first dimension, optionally values ​​of the second dimension and of the reflectivity, therefore means determining new values ​​minimizing the sum of the differences between the real intensities of the specific points of the group of points and their theoretical intensity determined at the end of operation 170. This amounts to considering that the section of the light beams associated with the specific points of the group of points presumed to belong to the object and estimated during operation 150 is closer to the section of the light beams actually reflected by the reflection surface of the object for the acquisition of these points.Now, this section is determined from the dimensions of the reflection surface of the object so that when the error of the cost function decreases, this means that the dimensions of the reflection surface are closer to the real dimensions of the reflection surface than when they were only estimated from the three-dimensional coordinates of the points of the group of points presumed to belong to the object. The method 100 therefore allows a more precise estimation of the value of the first dimension, possibly of the values ​​of the second dimension and / or of the reflectivity of the reflection surface, than of the values ​​obtained from the three-dimensional coordinates of points.

[0056] In the first, possibly, second and / or third examples, the determination of the iterative variation of the values ​​of the parameters of the model of the object to be corrected to minimize the error of the cost function is carried out by an optimization algorithm. In examples, the optimization algorithm chosen to determine the iterative variation of the values ​​of the parameters of the model of the object to be corrected may for example correspond to a Levenberg-Marquadt algorithm.

[0057] In examples in which several parameters of the object model are corrected by a minimization of the cost function, the cost function expresses the error to be minimized as a function of the values ​​of the parameters of the object model to be corrected and the minimization of the cost function comprises a calculation of the Jacobian matrix of the error to be minimized as a function of the parameters of the object model expressed in the cost function. The calculation of the Jacobian matrix of the error of the cost function as a function of the parameters of the object model to be corrected makes it possible to optimize the convergence of the optimization algorithm towards a minimization of the error by determining the relative impact of the variations of each parameter of the object model on the minimization of the error of the cost function.

[0058] In examples, and as illustrated by [Fig. 3], the method 100 may further comprise an operation 190 of determining a trajectory of the vehicle from the corrected estimate of the value of the first dimension determined during the operation 180. The trajectory of the vehicle may also be determined from the corrected estimate of the value of the second dimension determined during the operation 181. The trajectory of the vehicle may for example be determined so as to avoid the object whose dimensions were recalculated during process 100.

[0059] In examples, the plurality of points acquired by the lidar sensor 10 and on which the method is implemented is enriched by new acquisitions from the lidar sensor 10 and the method is implemented iteratively from the enriched plurality of points. In this case, the more points there are in a group of points presumed to belong to the same object, the more its dimensions can be determined precisely using the method according to the present disclosure.

[0060] Thus, the method 100 according to the present disclosure makes it possible to estimate the length of at least one side of the reflection surface of an object captured by the lidar sensor 10 in a precise manner and in particular in a more precise manner than a conventional estimation based solely on the three-dimensional coordinates of the points presumed to belong to the object, these dimensions possibly being under- or over-evaluated due to the resolution of the lidar sensor 10. In examples, the method 100 estimates the length of the two sides of the reflection surface of the object (width L and height h).These estimates of the dimensions of the object's reflection surface can in particular be used to determine a vehicle trajectory adapted to the precise dimensions of the object, which can make it possible to avoid situations in which a vehicle trajectory determined on the basis of underestimated dimensions of the object could be detrimental to the safety of the vehicle and road users.

Claims

Claims

1. Method for estimating at least one dimension belonging to a reflection surface (Sr) of an object, the method being implemented by a device (1) comprising at least one lidar sensor (10), a computer (11) and a memory (12), the memory (12) storing a model of the object comprising a first dimension and a second dimension of the reflection surface of the object and a reflectivity of the reflection surface of the object; the method comprising: - an acquisition (110) by the lidar sensor (10) of a set of points, each point being associated with three-dimensional coordinates in space and with an intensity value; - a selection (120) from the set of points of a group of points presumed to belong to the same object; - an estimation (130), from the three-dimensional coordinates of the points of the group of points, of the values of a first and a second dimension of the object corresponding respectively to the first and second dimensions of the reflection surface (Sr) of the model of the object; - an estimate (140), from the intensity values of the group of points presumed to belong to the object, of a reflectivity value of the object corresponding to the reflectivity of the reflection surface of the model of the object; then for at least one specific point of the group of points presumed to belong to the object: • an estimation (150) of a section of a specific light beam whose reflection by the reflection surface (Sr) of the object made it possible to obtain the specific point, from the estimated dimensions of the reflection surface of the object and a divergence of the light beam determined from a distance between the specific point and the lidar sensor; • an estimate (160) of a theoretical intensity value received by the lidar sensor (10) following the emission of the specific light beam, from the estimated section of the specific light beam and the estimated reflectivity value of the reflection surface; a determination (170) of an intensity error associated with the specific point from a difference between the estimated theoretical intensity value and the measured intensity value associated with the specific point following the emission of the specific light beam; and a correction (180) of the estimation of the value of the first dimension from the intensity error associated with at least one specific point.

2.

3.

4. Method according to the preceding claim, in which a correction of estimating the value of the first dimension from the intensity error associated with at least one specific point comprises minimizing an error of a cost function calculated from the intensity error associated with at least one specific point by iteratively varying the value of the first dimension. A method according to any preceding claim, further comprising: a correction (181) of the estimation of the value of the second dimension from the intensity error associated with at least one specific point, the correction comprising a minimization of an error of a cost function calculated from the intensity error associated with at least one specific point by iteratively varying the value of the second dimension. A method according to any preceding claim, further comprising: a correction (182) of the estimate of the reflectivity value from the intensity error associated with at least one specific point, the correction comprising a minimization of an error of a cost function calculated from the intensity error associated with at least one specific point by iteratively varying the reflectivity value.

5. A method according to any one of claims 2 to 4, wherein the error of the cost function corresponds to the sum of the intensity errors of the specific points for which the intensity error was determined.

6. A method according to any preceding claim, wherein the first dimension of the object model corresponds to a height (h) of the reflection surface (Sr) of the object and wherein the second dimension of the object model corresponds to a width (L) of the reflection surface of the object.

7. Method according to any one of the preceding claims, in which the theoretical intensity value received by the lidar sensor following the emission of the specific light beam is estimated from a function characterizing the intensity received by the lidar following the reflection of a light beam emitted by the lidar sensor on a surface as a function of: a. a distance between this surface and the lidar sensor, b. a reflectivity of this surface, and c. a section of the light beam reflected on this surface.

8. Method according to any one of the preceding claims, in which the plurality of points is enriched by new acquisitions of the lidar sensor (10) and the method is implemented iteratively from the enriched plurality of points.

9. Method according to any one of the preceding claims, further comprising a determination (190) of a trajectory of a vehicle (2) carrying the device from the corrected estimate of the value of the first dimension.

10. Device (1) comprising at least one lidar sensor (10), a computer (11) and a memory (12), the device (1) being configured to implement any one of the methods according to claims 1 to 9.

11. Vehicle (2) carrying a device (1) according to the preceding claim.

Citation Information

Patent Citations

  • Object detection based on lidar intensity

    EP3605137A1

  • Object recognition device and vehicle control system

    US20200301019A1