A design method and system for lidar test targets

Abstract

This invention belongs to the field of optical sensing signal processing and relates to a design method and system for lidar test targets. The method includes: obtaining a point cloud corresponding to the lidar parameters through simulation based on the lidar parameters of the lidar under test and the various reflective surfaces of the target model to be optimized; optimizing the performance discrimination of the point cloud using an evolutionary algorithm to obtain a test target with globally optimized performance discrimination for specific indicators. This invention considers the error factors of lidar and optimizes the test target, providing higher discrimination while reducing unexpected effects caused by factors such as multiple reflections. This invention can be used to design targets for testing lidar performance with different resolutions, ranging accuracy, etc., and by constraining the target shape, it can make it easier to manufacture.

Description

Technical Field

[0001] This invention relates to the field of optical sensing signal processing, and specifically to a design method and system for lidar test targets. Background Technology

[0002] LiDAR (Light Detection and Ranging) can create high-precision 3D maps of the environment through the emission and reflection of laser beams in multiple directions, and it does not rely on visible light, exhibiting more stable performance under different lighting conditions or inclement weather. In autonomous driving, by fusing with other sensors such as cameras, LiDAR enables vehicles to identify and distinguish road signs, pedestrians, obstacles, etc., and understand the surrounding environment, playing a crucial role. To ensure its safety, its measurement accuracy, resolution, and robustness to inclement weather need to be tested.

[0003] Indoor testing grounds can conveniently provide various environmental factors through facilities such as rain and snow chambers, but most existing test targets are still flat panels with different reflectivities. Since lidar generally has high accuracy, using ranging error as an indicator is insufficient to provide adequate performance differentiation under conditions of high reflectivity and high signal-to-noise ratio in lidar signals. Furthermore, suitable reflectivity is difficult to adjust. In addition, lidar ranging errors mainly originate from noise and timing errors in the photodetector, as well as errors caused by multiple reflections of the beam before returning to the photodetector, resulting in superimposed signals. While flat panel targets can produce some multiple reflections from reflective surfaces such as the ground, the impact on lidars with different performance characteristics is not controllable.

[0004] In summary, current lidar testing targets suffer from problems such as potentially low distinguishability between lidars with different performance levels and difficulty in control. Summary of the Invention

[0005] To address the problems described in the background art, this invention proposes a design method and system for lidar test targets. It optimizes the discrimination ability mentioned in the background art, providing higher discrimination in specific indicators such as the signal-to-noise ratio and resolution of the photodetector, and outputs the results in the form of ranging error, thereby improving testing efficiency and reducing the difficulty of adjusting the test site.

[0006] The technical solution adopted in this invention is as follows:

[0007] A method for designing a lidar test target includes the following steps:

[0008] Based on the lidar parameters of the lidar under test and the various reflective surfaces of the target model to be optimized, the point cloud corresponding to the lidar parameters is obtained through simulation.

[0009] Based on the performance discrimination of point cloud computing, and optimized using an evolutionary algorithm, a test target is obtained after global optimization of the performance discrimination of specific indicators.

[0010] Furthermore, the step of obtaining the point cloud corresponding to the lidar parameters through simulation based on the lidar parameters of the lidar under test and the various reflective surfaces of the target model to be optimized includes:

[0011] According to the lidar parameter θ={θ i Each lidar parameter θ in} i The set of emitted rays n is obtained from the field of view and angular resolution. i Thus, the reflecting surface {α} of the irradiated target model M is obtained. j The reflected light signal I(t) is calculated based on the optical path length, target reflectivity, and lidar parameters for single and multiple reflections.

[0012] The point cloud set P(θ) = {P} corresponding to each lidar parameter is obtained by using the reflected light signal I(t). i}, where P i Represents a lidar parameter θ i The corresponding point cloud.

[0013] Furthermore, the reflected light signal is calculated using the following formula:

[0014]

[0015] Where I(t) is the reflected light signal, I0 is the emitted signal, N(t) is the photodetector noise expressed as Gaussian white noise, r is the total length of the optical path, and c is the speed of light. σ represents the light intensity loss caused by both beam divergence and atmospheric scattering. ae η represents aerosol attenuation, φ represents light intensity loss due to reflection, and η represents the equivalent efficiency.

[0016] Furthermore, the calculation formula used for the point cloud computing performance differentiation is as follows:

[0017]

[0018] Where E(M,θ) represents the performance discrimination, P i Arranged in ascending order according to the values ​​of the lidar's measured indicators, f(M,P) i ) is the index function; D i Let γ represent the variance, β represent the weight of the variance term, and Δf(M,P) represent the penalty parameter for the exponential term. i ) represents adjacent f(M,P) i The difference between ).

[0019] Furthermore, the optimization using an evolutionary algorithm includes:

[0020] Constraints are added to the target model M, fixing the (x,y) coordinates of its vertices, and a 3D surface is formed by a triangular mesh. The target model M is represented by vertex vectors m = (m1 … m2). n ) represents the number of vertices in the model, where n is the number of vertices in the model, m1 … n Represents each vertex;

[0021] Generate a vertex vector population {m i}, and perform mutation and crossover, and calculate m in the mutated and crossover population. i Discrimination E(M) i After (θ), the fitness A is obtained. i ;

[0022] Select a certain proportion of dominant individuals, that is, those with high A... i m i The model is then replicated to the original population size to begin the next iteration, thereby optimizing the target model M.

[0023] Furthermore, the fitness A i Calculate using the following formula:

[0024] A i =E(M i ,θ)-P(m i )

[0025] Wherein, P(m) i ) for m i Range constraints, i.e., within m i The system provides a penalty when an element value is too high or too low, thereby limiting the production of models that are too large and difficult to process.

[0026] A design system for lidar test targets, comprising:

[0027] The point cloud computing module is used to obtain the point cloud corresponding to the lidar parameters through simulation based on the lidar parameters of the lidar under test and the various reflective surfaces of the target model to be optimized.

[0028] The test target optimization module is used to optimize the test target based on the performance discrimination of point cloud computing using an evolutionary algorithm, so as to obtain a test target with global optimization of the performance discrimination of specific indicators.

[0029] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0030] 1. This invention takes into account the error factors of lidar and optimizes the test target, which can reduce the unexpected impact caused by factors such as multiple reflections while providing higher discrimination.

[0031] 2. This invention can be used to design and test targets with different lidar performance characteristics such as resolution and ranging accuracy, and by constraining the target shape, it can make it easier to process. Attached Figure Description

[0032] Figure 1 This is a flowchart of the lidar test target design method of the present invention.

[0033] Figure 2 This is a schematic diagram of the distribution of vertices and edges of the laser radar test target in this invention.

[0034] Figure 3 It is the test target model obtained by optimizing the method of this invention.

[0035] Figure 4 It is the performance discrimination of the lidar on the test target and flat surface obtained by the method of this invention. Specific implementation methods

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below through specific steps and accompanying drawings. It should be understood that the specific steps described herein are for illustrative purposes only and are not intended to limit the invention.

[0037] This invention provides a method for designing a LiDAR test target. The method involves simulation under different LiDAR parameter settings to obtain a point cloud from the target to be optimized. The result is then input into a discrimination function as the fitness of an evolutionary algorithm for optimization, thereby obtaining a globally optimized test target with high discrimination for a specific metric. The process is as follows: Figure 1 As shown, the specific steps are as follows:

[0038] 1) Based on the lidar parameters θ={θ} obtained by sampling within the expected parameter range of the lidar under test i} and the various reflecting surfaces {α} of the target model M to be optimized j Point clouds are obtained through simulation.

[0039] Specifically, this can be obtained from the parameter θ of each lidar. i The set of emitted rays n is obtained from the field of view and angular resolution. i Thus, the irradiation of {α} is obtained. j The optical path for single and multiple reflections. Each reflected light signal can be calculated from the optical path length, target reflectivity, and lidar parameters:

[0040]

[0041] Where I(t) is the reflected light signal, I0 is the emitted signal, and N(t) is the photodetector noise expressed as Gaussian white noise. r is the total length of the optical path, and c is the speed of light. σ represents the light intensity loss caused by both beam divergence and atmospheric scattering. ae Let ρ represent aerosol attenuation, and η represent the light intensity loss due to reflection. η is the equivalent efficiency obtained from factors such as the optical receiver aperture and system efficiency. Since the parameters corresponding to η are difficult to obtain or predict, they can be calculated using the effective range and ranging accuracy of the lidar under different target reflectivities. Furthermore, the point cloud set P(θ) = {P...} corresponding to each lidar parameter can be obtained from the reflected light signal I(t). i}. Where P i Represents a lidar parameter θ i The corresponding point cloud. Taking a ToF lidar as an example, the peak value of I(t) can be detected, and the time difference between it and the peak value of the emitted signal I0 can be calculated, which is the flight distance of the emitted light in the direction corresponding to the signal, thereby obtaining its position in P. i The three-dimensional coordinates of the corresponding single point.

[0042] 2) Calculate the performance discrimination from the point cloud P(θ).

[0043] Specifically, the performance discrimination is described using the following function E(M,θ):

[0044]

[0045] Among them, P i Arranged in ascending order according to the values ​​of the lidar's measured indicators, f(M,P) i ) is an index function, such as point density, point cloud P i The distance between the target model M and the target model, i.e., the ranging error, etc.; D i Let γ represent the variance, β represent the weight of the variance term, and β represent the penalty parameter for the exponential term. That is, E(M,θ) primarily uses variance to represent discrimination, but to avoid E(M,θ) being affected by a few outliers, or f(M,P)... i The relationship between the performance of M and P lidar is not monotonic, making the actual test results difficult to interpret. A second exponential term has also been added. Among them, Δf(M,P) i ) represents adjacent f(M,P) i The difference between ) is such that f(M,P) is obtained without being significantly affected by outliers. i Add penalties for cases where the ascending order is not satisfied or the difference between two items is too small.

[0046] 3) Optimize the target model M.

[0047] To reduce the complexity of M and avoid obtaining a physically unrealizable target model, constraints need to be added to it before optimization. One method is to fix the (x,y) coordinates of the vertices of M, and then use constraints such as... Figure 2 The triangular mesh shown forms a three-dimensional surface, then M can be represented by vertex vectors m = (m1 … m2). n ) represents the number of vertices in the model. Where n is the number of vertices in the model, m1 n Represents each vertex.

[0048] Since E(M,θ) is not differentiable and the solution space corresponding to m is very large, an evolutionary algorithm is used to optimize it. That is, a vertex vector population {m} is generated. i}, where m i Let m represent the vector of the i-th vertex, and then mutate it (i.e., randomly modify some individuals m). i The corresponding vertex vector is generally in m i (Superimposed with random noise), crossover (i.e., from m) i m j Calculating new individuals is generally done for m. i m j Perform a weighted average or in m i m j (Each element is extracted from the population to form a new vertex vector). In the modified population obtained through mutation and crossover, m is calculated according to steps 1) and 2). i Discrimination E(M) i After (θ), its fitness is obtained:

[0049] A i =E(M i ,θ)-P(m i (3)

[0050] Among them, M i Represents the vertex vector m i The corresponding target model, P(m) i ) for m i Range constraints, i.e., within m i A penalty is applied when an element's value is too high or too low, thus limiting the generation of models that are too large and difficult to process. A certain proportion of dominant individuals, i.e., those with high A... i m i The model is then replicated to the original population size to begin the next iteration, thereby optimizing the target model M. Example:

[0051] This embodiment uses lidar parameters with a root mean square error of 1-2 cm, a field of view of 120°×20°, and an angular resolution of 0.1°, at a distance of 20m from the target, to optimize the discrimination of ranging accuracy. Details are as follows:

[0052] 1) Based on θ={θ i} and the various reflecting surfaces {α} of the target model M to be optimized j Point clouds are obtained through simulation.

[0053] The set of emitted rays n is obtained from the field of view and angular resolution of the lidar. i Thus, the irradiation of {α} is obtained. j The optical path for single and multiple reflections. Each reflected light signal can be calculated from the optical path length, target reflectivity, and lidar parameters:

[0054]

[0055] Where I0 is the transmitted signal, N(t) is the photodetector noise expressed as Gaussian white noise, and r is the total optical path length. ρ represents the light intensity loss caused by beam divergence and atmospheric scattering, while ρ represents the light intensity loss caused by reflection. Since reflectivity is difficult to adjust during manufacturing, it is set to 10% diffuse reflectivity, commonly used in autonomous driving calibration. η is the equivalent efficiency obtained from factors such as the optical receiver aperture and system efficiency. By adjusting η, the ranging errors of the five parameters sampled on a flat plate with the same reflectivity are 9mm, 11mm, 14mm, 17mm, and 19mm, respectively. Furthermore, the point cloud P(θ) = {P...} corresponding to each lidar can be obtained through the returned signal. i}

[0056] 2) Calculate the performance discrimination from the point cloud P(θ).

[0057] Specifically, the performance discrimination is described using the following function E(M,θ):

[0058]

[0059] Among them, P i Sort by the root mean square error of the lidar ranging in ascending order, D i The variance is Δf(M,P). In the second term, Δf(M,P) represents the variance. i ) represents adjacent f(M,P) i The difference is such that f(M,P) is obtained without being significantly affected by outliers. i A penalty is added if the ascending order is not satisfied or if the difference between two items is too small. f(M,P) i ) represents the root mean square error (RMSE) of the distance measurement, i.e., P i The root mean square of the distance from each point to M:

[0060]

[0061] Where d(M,p) is P iThe shortest distance from point p to M. In practice, the variance term weight γ can be determined according to the selected f(M,P) value. i The magnitudes of the variance and exponent terms are adjusted to bring them to the same order of magnitude. β can be set such that βΔf(M,P) i ) in 1 to 10 1 Initial values ​​of order of magnitude. Both γ and β can be calculated based on whether the optimized model M satisfies f(M,P). i Adjust the order of items in ascending order if the difference between the two items is large enough.

[0062] 3) Optimize the target model M.

[0063] To reduce the complexity of M and avoid obtaining a physically unrealizable target model, constraints need to be added before optimization. Here, the projections of each vertex of M onto the xOy plane are fixed on an 8×8 square grid with a side length of 0.2m, meaning that the (x,y) coordinates of its vertices remain unchanged, and are determined by... Figure 2 The triangular mesh shown connects the edges to form continuous triangular faces, thus allowing it to be automatically recognized as a three-dimensional surface. M can then be represented by the vertex vector m = (m1 … m2). n ) represents the number of vertices in the model. Where n = 64 is the number of vertices in the model

[0064] The model is optimized using an evolutionary algorithm. This involves generating a vertex vector population {m}. i}, and perform mutation and crossover. Mutation occurs in m i A Gaussian noise is superimposed on top. The crossover is a pair of m. i m j Perform a weighted average with random weights.

[0065] In the modified population, calculate m according to 1) and 2). i Discrimination E(M) i After (θ), its fitness is obtained:

[0066] A i =E(M i ,θ)-P(m i (7)

[0067] Where P(m) i ) for m i The range constraint is set at m. i A penalty is applied when element values ​​do not fall within the range [0, 0.5m], thus limiting the generation of excessively large and difficult-to-manufacture models. A certain proportion of dominant individuals, i.e., those with high A... i m i For example, the top 10% of m can be selected. i The model is then replicated to the original population size to begin the next iteration, thereby optimizing the target model M. The optimization results are as follows: Figure 3 As shown.

[0068] To verify its effectiveness, using the same lidar parameters, a flat plate with a 10% diffuse reflectivity and perpendicular to the lidar's direction was placed 20m away from the lidar. To eliminate the influence of different target viewing angles, the flat plate and the optimized target model had the same dimensions, 1.4m × 1.4m. The root mean square error of ranging obtained by the flat plate and the optimized target model under different lidar parameters was calculated according to equation (6), and the results are as follows. Figure 4 As shown, the difference in ranging error between the optimized target and different lidar parameters is 5mm to 12mm, which is higher than the 2mm to 3mm difference of the flat plate, indicating higher discrimination.

[0069] Another embodiment of the present invention provides a design system for lidar test targets, comprising:

[0070] The point cloud computing module is used to obtain the point cloud corresponding to the lidar parameters through simulation based on the lidar parameters of the lidar under test and the various reflective surfaces of the target model to be optimized.

[0071] The test target optimization module is used to optimize the test target based on the performance discrimination of point cloud computing using an evolutionary algorithm, so as to obtain a test target with global optimization of the performance discrimination of specific indicators.

[0072] The above division of modules is merely illustrative. In practical applications, the functions described above can be assigned to different functional modules as needed to complete all or part of the functions described in the aforementioned method. The specific working process of each module can be found in the corresponding process in the aforementioned method embodiments, and will not be repeated here.

[0073] Another embodiment of the present invention provides a computer device (computer, server, smartphone, etc.) including a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for performing the steps of the method of the present invention.

[0074] Another embodiment of the present invention provides a computer-readable storage medium (such as ROM / RAM, disk, optical disk) storing a computer program that, when executed by a computer, implements the various steps of the method of the present invention.

[0075] The laser radar test target design method of the present invention has been described in detail above, but it is obvious that the specific implementation of the present invention is not limited thereto. For those skilled in the art, various obvious modifications made to the method without departing from the spirit and scope of the claims are within the protection scope of the present invention.

Claims

1. A method for designing a lidar test target, characterized in that, Includes the following steps: Based on the lidar parameters of the lidar under test and the various reflective surfaces of the target model to be optimized, the point cloud corresponding to the lidar parameters is obtained through simulation. Based on the performance discrimination of point cloud computing, and optimized using an evolutionary algorithm, a test target is obtained after global optimization of the performance discrimination of specific indicators.

2. The method according to claim 1, characterized in that, The step of obtaining the point cloud corresponding to the lidar parameters through simulation based on the lidar parameters of the lidar under test and the various reflective surfaces of the target model to be optimized includes: According to the lidar parameter θ={θ i Each lidar parameter θ in} i The set of emitted rays n is obtained from the field of view and angular resolution. i Thus, the reflecting surface {α} of the irradiated target model M is obtained. j The reflected light signal I(t) is calculated based on the optical path length, target reflectivity, and lidar parameters for single and multiple reflections. The point cloud set P(θ) = {P} corresponding to each lidar parameter is obtained by using the reflected light signal I(t). i }, where P i Represents a lidar parameter θ i The corresponding point cloud.

3. The method according to claim 2, characterized in that, The reflected light signal is calculated using the following formula: Where I(t) is the reflected light signal, I0 is the emitted signal, N(t) is the photodetector noise expressed as Gaussian white noise, r is the total length of the optical path, and c is the speed of light. σ represents the light intensity loss caused by both beam divergence and atmospheric scattering. ae ρ represents aerosol attenuation, η represents light intensity loss due to reflection, and η represents the equivalent efficiency.

4. The method according to claim 2, characterized in that, The calculation formula used for the point cloud computing performance differentiation is as follows: Where E(M, θ) represents the performance discrimination, P i Arranged in ascending order according to the values ​​of the lidar's measured indicators, f(M, P) i ) is the index function; D i Let γ represent the variance, β represent the weight of the variance term, and Δf(M, P) represent the penalty parameter for the exponential term. i ) are adjacent f(M, P) i The difference between ).

5. The method according to claim 4, characterized in that, The optimization using an evolutionary algorithm includes: Constraints are added to the target model M, fixing the (x, y) coordinates of its vertices, and a 3D surface is formed by a triangular mesh. The target model M is represented by vertex vectors m = (m1…m2). n ) represents the number of vertices in the model, where n is the number of vertices in the model, m1… n Represents each vertex; Generate a vertex vector population {m i }, and perform mutation and crossover, and calculate m in the mutated and crossover population. i Discrimination E(M) i After (θ), the fitness A is obtained. i ; Select a certain proportion of dominant individuals, that is, those with high A... i m i The model is then replicated to the original population size to begin the next iteration, thereby optimizing the target model M.

6. The method according to claim 5, characterized in that, The fitness A i Calculate using the following formula: A i =E(M i ,θ)-P(m i ) Wherein, P(m) i ) for m i Range constraints, i.e., within m i The system provides a penalty when an element value is too high or too low, thereby limiting the production of models that are too large and difficult to process.

7. The method according to claim 5, characterized in that, The (x, y) coordinates of the vertices of the fixed target model M are obtained by fixing the projection of each vertex of the target model M onto an 8×8 square grid with a side length of 0.2m, so that the (x, y) coordinates of the vertices remain unchanged.

8. A design system for laser radar test targets, characterized in that, include: The point cloud computing module is used to obtain the point cloud corresponding to the lidar parameters through simulation based on the lidar parameters of the lidar under test and the various reflective surfaces of the target model to be optimized. The test target optimization module is used to optimize the test target based on the performance discrimination of point cloud computing using an evolutionary algorithm, so as to obtain a test target with global optimization of the performance discrimination of specific indicators.

9. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for performing the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a computer, implements the method according to any one of claims 1 to 7.

Images