Method, computer program and device for determining test limits for testing the suitability of optical components for use in an overall optical system

Abstract

The present invention relates to a method, to a computer program with instructions and to a device for determining test limits for testing the suitability of optical components for use in an overall optical system used by an algorithm. In a first step, distributions of the optical quality of the individual optical components are determined (10) on the basis of optical measurements in the amplitude space, said distributions being subject to tolerances. In addition, a combination model is calibrated (11) on the basis of measurements of the optical quality of actual overall optical systems. A distribution of the optical quality of the overall optical system is determined (12) by combining the optical components using the calibrated combination model, said optical components having tolerances. Functional limits of a task model of the algorithm are determined (13) depending on the optical quality of the overall system. Finally, test limits are determined (14) for at least one of the optical components by mapping the functional limits of the task model to the individual contributions of the optical components.

Description

[0001] Description

[0002] Method, computer program and device for determining test limits for testing the suitability of optical components for use in a complete optical system

[0003] The present invention relates to a method, a computer program with instructions and a device for determining test limits for testing the suitability of optical components for use in an overall optical system, which is used by an algorithm.

[0004] The optical quality of windshields is relevant for the use of camera systems behind the glass, for example, for use in driver assistance systems or even self-driving vehicles. For instance, a windshield must have sufficient quality so that, for example, the traffic sign recognition of the driver assistance system functions as specified throughout the vehicle's lifespan.

[0005] There is increasing cost pressure in the testing of windshields. Because existing solutions are not sufficiently informative, good components may be classified as defective and destroyed. Similarly, defective components might not be identified as such. Since this could lead to a safety issue, it must be ensured that no defective components are installed and delivered. This is particularly relevant for autonomous driving, where there is no human operator to compensate for a sudden malfunction. Therefore, existing tests are subject to very tight limits, typically imposed by the algorithm suppliers, resulting in increased scrap and significant costs.

[0006] Against this background, US 2022 / 0412897 A1 describes a method for measuring the optical quality of a region of a vehicle windshield, wherein the region is intended to be positioned in the optical path of an image capture device. In the measurement method, a beam of light rays is emitted towards the region by means of an emitter. Using a wavefront analyzer, the wavefront of the light rays transmitted by the region is analyzed, thereby creating a wavefront defect map. Based on the wavefront defect map, a map of the optical defects present in the region is determined. DE 102020215417 A1 describes a method for measuring the influence of a transparent windshield, in which a displacement field induced by the windshield is determined. In a first step, an initial image of a textured surface without the transparent windshield is acquired.In a second step, a second image of the textured surface with the transparent disc is taken. In a third step, the displacement field is determined by analyzing the two images using an optical flow method.

[0007] US 2023 / 0377245 A1 describes a method for simulating the effects of the refractive power of a windshield on the image quality of a digital imaging device. The method is based on a modified stochastic ray tracing technique.

[0008] US 5179273 A describes a method for testing the performance of adaptive optics. The method uses localized wavefront power analysis to determine where and with what force the adaptive optics must be deformed to improve its performance.

[0009] For testing windshields, a solution is needed that allows for a reliable assessment of whether the windshield is suitable for use in front of cameras. To ensure a reliable assessment, it is necessary to determine how well the metric used correlates with the performance of the function. This question therefore concerns the relationship between optical quality and function. Regarding the evaluation of whether the windshield is suitable, it is necessary to determine the appropriate test limits when the metric correlates well. This question therefore concerns the process of determining the functional limits.

[0010] In addition to these two optical and metrological challenges, there is another problem, also based on a technical principle. The windshield, together with the camera lens, forms a new optical system that behaves differently than its individual components. In other words, the transfer functions of the individual components are not multiplicative, but rather combine non-linearly. Since the final optical quality is only achieved through the assembly of the windshield and camera in the vehicle, it cannot be directly determined from the individual parts. This leads to an organizational and commercial problem. The entire automotive supply chain is designed so that testing is carried out at the component manufacturer's site—that is, one test at the windshield manufacturer's site and another at the camera manufacturer's site.An inspection at the end of the production line at the vehicle manufacturer cannot be carried out because a negative result from the system test—i.e., that this windshield and this camera together fall below the test limits—would require at least the camera to be replaced. Since this would occur when almost the entire vehicle is already built, this is not feasible.

[0011] Two measurement methods are currently used to test the optical quality of windshields, based on refractive power and modulation transfer function.

[0012] Refractive power measurements have been used for decades to examine the influence of the windshield on human perception. The sole purpose is to determine whether a person can see through the windshield without distortion. Refractive power is measured horizontally and vertically, originally across the entire width and height of the windshield with low spatial resolution, using collimated light.

[0013] Driver assistance cameras have now become standard, mounted directly behind the windshield in front of the rearview mirror. These cameras look through a trapezoidal window, which is covered with a black print for both aesthetic and technical reasons. Until now, existing measurement technology has been used within this viewing window without questioning its functionality.

[0014] Within the aforementioned window, a much higher spatial resolution of the refractive power is desired, which is why refractive power measurements across the entire windshield are typically not feasible. Therefore, the refractive power measurement within the camera window was adapted. For this purpose, a refractive power measurement method was developed that inherently accounts for the variation in field angles within the camera lens's field of view. An identical camera lens is inserted into the measuring device, and the distortion of the entire system is determined using the background-oriented schlieren method. The high-resolution measurement information can then be used for the numerical post-hoc correction of the image distortion.

[0015] However, there are two fundamental problems with refractive power measurement. Firstly, refractive power is not a suitable metric. It has been shown that refractive power correlates very poorly with the performance of a machine learning algorithm. Therefore, refractive power cannot be used to reliably predict whether and what kind of influence the quality of the windshield will have on the algorithm. Consequently, it cannot be used to determine the test limits. Secondly, certain relevant optical information, especially oblique astigmatism, is fundamentally missing from the refractive power measurement. The windshield can cause a certain degree of blurring that cannot be compensated for by the current refractive power measurement standard. Therefore, refractive power is unsuitable for evaluating the quality of a windshield for the aforementioned camera systems.

[0016] The modulation transfer function (MFT) is a quantity derived from systems theory used to evaluate the "sharpness" of a camera. According to systems theory, it is the transfer function in the frequency domain of the lens as an optical system, and in this sense, at least abstractly, it could also be used to evaluate the quality of a windshield. However, a windshield is not an imaging system, which is a prerequisite according to the definition of Fourier optics. Nevertheless, the MFT has long been an established measurement parameter for camera systems in the automotive sector.

[0017] For the past few years, the modulation transfer function (MTF) has been used to assess the quality of windshields. Initial projects are underway that define specific test limits for the MTF. The actual test metric is the drop in the MTF, which is tolerated only when the windshield is mounted in front of the camera. In other words, the camera is measured separately. When mounted behind the windshield, the camera's MTF must not drop by more than a defined limit.

[0018] The camera and the windshield are not independent systems that can be evaluated separately using the modulation transfer function in a component test, and therefore the system performance of the combined system cannot be predicted. This is because the combination of windshield and camera constitutes a new optical system. This becomes clear, for example, when the camera image can be sharper if the windshield is mounted in front of the camera. An object of the invention is to provide suitable solutions for determining test limits for testing the suitability of optical components for use in a complete optical system.

[0019] This problem is solved by a method having the features of claim 1, by a computer program with instructions according to claim 14, and by a device having the features of claim 15. Preferred embodiments of the invention are the subject of the dependent claims.

[0020] According to a first aspect of the invention, a method for determining test limits for testing the suitability of optical components for use in an overall optical system, which is used by an algorithm, comprises the steps: - Determining tolerance distributions of the optical quality of the individual optical components based on optical measurements in the amplitude space;

[0021] - Determining a distribution of the optical quality of the overall optical system by combining the tolerated optical components using a combination model calibrated on the basis of measurements of the optical quality of real overall optical systems;

[0022] - Determining functional limits of a task model of the algorithm depending on the optical quality of the overall system; and

[0023] - Determining test limits for at least one of the optical components by mapping the functional limits of the task model onto the individual contributions of the optical components.

[0024] According to another aspect of the invention, a computer program contains instructions which, when executed by a computer, cause the computer to perform the following steps for determining test limits for testing the suitability of optical components for use in an overall optical system utilized by an algorithm:

[0025] - Determining tolerance-related distributions of the optical quality of individual optical components based on optical measurements in amplitude space;

[0026] - Determining a distribution of the optical quality of the overall optical system by combining the tolerated optical components using a combination model calibrated on the basis of measurements of the optical quality of real overall optical systems;

[0027] - Determining functional limits of a task model of the algorithm as a function of the optical quality of the overall system; and - Determining test limits for at least one of the optical components by mapping the functional limits of the task model onto the individual contributions of the optical components.

[0028] The term "computer" is to be understood broadly. In particular, it also includes workstations, distributed systems, and other processor-based data processing devices. Furthermore, the individual steps are not necessarily performed directly by the computer. It is equally possible that the computer controls or relies on external components to carry out individual steps.

[0029] The computer program can, for example, be made available for electronic retrieval or be stored on a computer-readable storage medium.

[0030] According to a further aspect of the invention, a device for determining test limits for testing the suitability of optical components for use in a complete optical system used by an algorithm comprises a memory containing instructions and a processor, wherein the processor is configured, upon execution of the instructions, to perform the following steps for determining test limits for testing the suitability of optical components for use in a complete optical system used by an algorithm: - Determining tolerance-related distributions of the optical quality of the individual optical components based on optical measurements in the amplitude space;

[0031] - Determining a distribution of the optical quality of the overall optical system by combining the tolerated optical components using a combination model calibrated on the basis of measurements of the optical quality of real overall optical systems;

[0032] - Determining functional limits of a task model of the algorithm depending on the optical quality of the overall system; and

[0033] - Determining test limits for at least one of the optical components by mapping the functional limits of the task model onto the individual contributions of the optical components.

[0034] In the solution according to the invention, test limits are to be determined for the individual components of a complete optical system, even though only the assembled system possesses the actual optical quality. A relationship is thereby established between the functional limits of the task-solving algorithm and this optical quality of the complete system.

[0035] According to the invention, optical quality is measured in the amplitude domain. First, the tolerances of the individual components are determined metrologically using a statistically significant number of components. From these tolerance-related distributions, the toleranced components can be combined using a calibrated combination model. In this way, a distribution of the optical quality of the overall optical system can be determined. For calibration purposes, the distribution of optical quality of the overall optical system resulting from the combination model can be compared with a distribution of optical quality obtained by measuring a statistically significant number of real optical systems. This allows for a comparison between the combination model and reality.

[0036] The optical quality of the overall optical system is used in algorithm development to determine the functional limits of the task model. This establishes a relationship between the parameterized tolerances of the individual components and their influence on the task model within the overall system. The functional limits of the task model are then determined based on the optical system quality and can subsequently be mapped to the individual contributions of the optical components.

[0037] To determine the inspection limits, a correlation between the detection accuracy of the algorithms and a scalar optical metric can be identified. A scalar optical metric could be, for example, the optical informative gain (OIG), described in the article by DW Wolf et al.: "Sensitivity analysis of Al-based algorithms for autonomous driving on optical wavefront aberrations induced by the windshield", arXiv:2308.11711, or the Strehl ratio, etc. Based on this bijective relationship, inspection limits can now be defined, preferably by the manufacturer of the overall optical system, since the inspection limits may quantitatively depend on the basic architecture of the technical system. The described process thus enables a reliable release of the overall optical system with regard to optical quality.

[0038] The solution according to the invention enables both end-of-line testing and individual component testing at the suppliers of the optical components, even though a nonlinear system is used. Due to the nonlinear dependence determined by the neural surrogate model between the test limits of the individual components and the test limits of the effective overall optical system, the proportion of optical components incorrectly classified as rejects can be significantly reduced.

[0039] The described solution can also be extended to the aftermarket, for example, to test a newly installed lens that has been previously measured using wavefront measurements. This ensures that the optical quality of the overall system meets the required standards. For this to work, the retrofitter must have a testing system installed that is equivalent to the measuring system used by the original manufacturer of the complete optical system.

[0040] According to one aspect of the invention, the combination model for the tolerated optical components is based on a statistical method. This can, for example, utilize Latin hypercube sampling or Monte Carlo sampling. Both approaches are well suited for realizing a combination model.

[0041] According to one aspect of the invention, the optical quality of the entire system is used as a perturbation model in the development of the algorithm to determine the functional limits of the task model of the algorithm as a function of the optical quality of the entire system.

[0042] According to one aspect of the invention, the optical measurements are wavefront measurements. Wave optics is the fundamental description of light as classical, i.e., non-quantum mechanical, electromagnetic radiation. Wave optics uses the wavefronts of light to describe optical systems where the light is modeled in the form of plane or spherical waves. These wavefronts can be detected using established measurement methods, in particular, for example, a Shack-Hartmann sensor.

[0043] According to one aspect of the invention, the test limits relate to wavefront measurements on the at least one optical component. For example, the test limits can define permissible value ranges for coefficients determined from the wavefront measurements.

[0044] According to one aspect of the invention, the measured optical qualities of the optical components and the overall optical system are parameterized using a decomposition into an orthonormal functional basis for describing the measured wavefronts. For example, the measured optical qualities of the optical components and the overall optical system can be parameterized by Zernike coefficients or by Legendre coefficients. Assuming a circular aperture diaphragm, the effective wavefront aberrations are expediently parameterized by Zernike coefficients. For a rectangular aperture diaphragm, parameterization by Legendre coefficients is more suitable.

[0045] According to one aspect of the invention, the overall optical system is part of a perception system. This perception system can, in particular, serve as an assistance function or a driving function in a means of transportation. In such perception systems, it is essential to ensure that no inadequate components are installed and delivered in order to avoid safety problems.

[0046] According to one aspect of the invention, the optical components are a disc and a camera lens. Complete optical systems consisting of a disc and a camera are increasingly being installed in means of transportation, particularly in motor vehicles.

[0047] According to one aspect of the invention, the algorithm is implemented as a neural network. In particular, the neural network can be trained for segmentation. Typical task-solving neural networks are used, for example, in instance segmentation, semantic segmentation, panoptic segmentation, object recognition, or object classification.

[0048] Further features of the present invention will become apparent from the following description and the attached claims in conjunction with the figures.

[0049] Fig. 1 schematically shows a perception system of a means of transportation;

[0050] Fig. 2 schematically shows a method for determining test limits for testing the suitability of optical components for use in a complete optical system;

[0051] Fig. 3 shows an embodiment of a device for determining test limits for testing the suitability of optical components for use in a complete optical system; and

[0052] Fig. 4 shows an embodiment of a neural network suitable for implementing a solution according to the invention. For a better understanding of the principles of the present invention, embodiments of the invention are explained in more detail below with reference to the figures. It is understood that the invention is not limited to these embodiments and that the described features can also be combined or modified without departing from the scope of protection of the invention as defined in the appended claims.

[0053] Fig. 1 schematically shows a perception system 1 of a means of transport 40. In this example, the perception system 1 comprises an optical system 2 and a data processing system 5, which performs an assistance or driving function of the means of transport 40. The optical system 2 comprises a first optical component 3, here a windshield, and a second optical component 4, here a camera lens. The images provided by the camera lens can be used by the data processing system 5, for example, for lane detection or the detection of pedestrians or other road users.

[0054] Figure 2 schematically shows a method for determining test limits for verifying the suitability of optical components for use in a complete optical system, which is used by an algorithm. The complete optical system can be part of a perception system, for example, for an assistance function or driving function in a means of transportation. In a first step, tolerance distributions of the optical quality of the individual optical components are determined based on optical measurements in the amplitude space. 10. In addition, a combination model is calibrated based on measurements of the optical quality of real complete optical systems. 11. The combination model is preferably based on a statistical method and can, for example, use Latin hypercube sampling or Monte Carlo sampling. The measurements can, in particular, be wavefront measurements.The measured optical qualities of the optical components and the overall optical system are parameterized using a decomposition into an orthonormal functional basis to describe the measured wavefronts, e.g., by Zernike coefficients or Legendre coefficients. By combining the toleranced optical components using the calibrated combination model, a distribution of the optical quality of the overall optical system is determined. Depending on the optical quality of the overall system, functional limits of a task model of the algorithm are determined. For this purpose, the optical quality of the overall system can be used, for example, as a perturbation model in the development of the algorithm. Finally, by mapping the functional limits of the task model onto the individual contributions of the optical components, test limits for at least one of the optical components are determined.

[0055] Fig. 3 shows a simplified schematic representation of an embodiment of a device 30 for determining test limits for testing the suitability of optical components for use in a complete optical system. The device 30 comprises a processor 32 and a memory 31. For example, the device 30 is a computer, a workstation, or a distributed system. Instructions are stored in the memory 31 which, when executed by the processor 32, cause the device 30 to perform the steps according to one of the described methods. The instructions stored in the memory 31 thus embody a program executable by the processor 32, which implements the method according to the invention.The device 30 has an input 33 for receiving information, in particular measurement data MD1, MD2 from optical measurements in the amplitude range for the optical components, and measurement data MDG from optical measurements in the amplitude range for the overall optical system. Data generated by the processor 32, for example, determined test limits PG, are provided via an output 34. Furthermore, data can be stored in memory 31. The input 33 and the output 34 can be combined into a bidirectional interface.

[0056] The processor 32 can comprise one or more processor units, such as microprocessors, digital signal processors, or combinations thereof.

[0057] Memory 31 can have both volatile and non-volatile memory areas and can include a wide variety of storage devices and storage media, such as hard disks, optical storage media or semiconductor memory.

[0058] A preferred embodiment of the invention will now be explained. The optical system under consideration includes, by way of example, a windshield and a camera lens. The objective function is also, by way of example, semantic segmentation.

[0059] Fig. 4 shows an embodiment of a neural network N suitable for implementing a solution according to the invention. This is, by way of example, a multifunctional neural network N that also performs semantic segmentation. Of course, other objective functions can be implemented instead of semantic segmentation. In the example shown, the neural network N is a U-NET-based convolutional network for estimating the effective wavefront aberrations W. pof the overall optical system, consisting of the windshield and camera lens. The neural network N comprises an encoder ENC, a bottleneck BN, a coupled decoder head consisting of a decoder DEC and a restorer RES, as well as skip connections SC for the features of the different resolution levels. Additionally, the architecture includes a residual encoder RN for identifying the second-order Zernike coefficients (Z3, Z4, Z5). The input for the neural network is a batch Bd of degraded images.

[0060] The DEC decoder for semantic segmentation uses pre-trained weights from training on undistorted raw images without optical aberrations. The RES restorer, designed to restore undistorted images from distorted ones, is added in parallel. The RN residual encoder, for identifying second-order Zernike coefficients, consists of several ResNet cells that calculate the difference between the neural network input N (batch Bd of degraded images) and the output of the RES restorer (batch B). w recovered images, process, i.e., a batch B r of residues.

[0061] The input format for the neural network N, i.e., the batch Bd of degraded images, is assumed to be as follows: number of images times image height in pixels times image width in pixels times number of color channels. Several hyperparameters can be adjusted, if necessary, before the model can be trained:

[0062] - N_classes: Number of classes for the DEC decoder for semantic segmentation - conv2D_kernel: Size of the convolution kernel tensor in pixels

[0063] - conv2D_increment: Increment of the convolution operation in pixels

[0064] - regu_strength: Weight of the regularization term

[0065] - Iearning_rate: Weighting factor for the total loss function

[0066] - batch_size: Number of processed images before the trainable variables are updated

[0067] - class_weights: Weighting factors for the individual classes based on their occurrence in the dataset

[0068] - Ioss_focus: Exponent that forces a concentration of the learning process on classes that are difficult to learn.

[0069] - Ioss_balance_restorer: Weighting factor for the loss of the restorer RES (L1 loss) - Ioss_balance_analyzer: Weighting factor for the loss of the analyzer, i.e., the residual encoder RN (L2 loss). The encoder ENC consists of five encoder blocks, with the number of filters doubling in each subsequent block. Each encoder block consists of two convolution layers, followed by a batch normalization layer and a max-pooling layer for downsampling. After the batch normalization layer, the computation graph is split into two branches to pass information to the decoder DEC, thus mitigating the vanishing gradient problem.

[0070] The Bottleneck BN consists of two convolution layers and subsequent batch normalization layers.

[0071] The restorer RES consists of five transposed convolution blocks, with the number of filters halved in each subsequent block. Each transposed convolution block comprises a transposed convolution layer, followed by a batch normalization layer and a merge node to incorporate, with high reliability, the information provided by the skip connection SC of the corresponding encoder block. The chained tensor is then smoothed by two convolution layers with a step size of one to preserve the dimensions.

[0072] The residual encoder RN, used to identify Zernike coefficients, consists of five ResNet cells for classifying aberrations based on second-order Zernike coefficients. Each ResNet cell comprises two convolution layers, two batch normalization layers, and a concatenation layer that combines the convolutional signal with the input signal to mitigate the vanishing gradient problem. Finally, a Max pooling layer is employed for downsampling. After the ResNet cells, the signal is smoothed and divided into five dense layers with batch normalization and dropout for regularization.

[0073] The DEC decoder for semantic segmentation consists of five transposed convolution blocks, with the number of filters halved in each subsequent block. These transposed convolution blocks for semantic segmentation are structured similarly to the transposed convolution blocks of the RES restorer.

[0074] The network N is trained using a fitted loss function. The loss function consists of three components. The first component is the negative log-likelihood for semantic segmentation, which corresponds to the cross-entropy. The second term quantifies the deviation between the unbiased image and the recovered image using the L1 norm. The third term quantifies the deviation between the predicted vector of Zernike coefficients and the vector of basic truth using the L2 norm. All components are summed, taking into account a weighting factor for hypertuning.

[0075] For training the neural network N, a physically realistic degradation model F based on Fourier optics is used. Such a model F requires wavefront measurement results W. gregarding a representative number of windshields and camera lenses that correspond to the project status. With this model F, the training data of the base network is now degraded, i.e., a batch B c undistorted images from a set AT of images taken without a windshield. To strengthen the robustness of the N network against optical aberrations, the degraded training data can be used for transfer learning. The components required for training are enclosed in a dashed box in Fig. 4.

[0076] To obtain a batch Bd of degraded image data, both the lens aberrations and the windshield aberrations must be considered. Measurement results from a wavefront measurement system, e.g., based on the background-oriented schlieren method, serve as the basis for the overall aberrations to be learned. In the illustrated embodiment, the aberrations of the individual components are incorporated by a Fourier optical model F. The batch of degraded image data Bd consists of a whole ensemble of different lenses and windshields, so that the definition range of the input mesh N is sufficiently covered.Once a specific lens has been measured and the derived optical system limits are known, the necessary disk quality can be empirically determined, since the non-linearity is compensated for by the information capacity of the surrogate model, consisting of the coupled decoder head and the ResNet encoder RN for the prediction of the Zernike coefficients.

[0077] The predicted total aberrations W pThe predicted Zernike coefficients, as shown here, serve as an additional input signal in the illustrated example to provide calibrated confidences K during operation. The input for the neural network is again a batch Bd of degraded images, but these now originate from a set AB of images taken with the windshield visible. The estimated wavefront aberrations of the overall optical system allow for the calibration of the network's predicted confidences N. The class-wise confidences can be used to determine the prediction uncertainty. This measure of uncertainty is important for decision-making regarding the subsequent action of the automated driving system, e.g., braking, swerving, etc. To determine the calibrated confidences K, the logit tensor LT output by the decoder DEC and the predicted total aberrations W are used. pThe data is fed into a calibration block P, which determines calibration factors for the predicted confidences using parameterized temperature scaling. The calibrated confidences K obtained in this way are transferred to a segmentation map S, which is output to downstream systems.

[0078] The implementation described above assumes a UNET as the basic architecture; however, the approach can also be applied to other network architectures. The proposed architecture can be constructed in a modified form with the same objective. Generally, the network architecture of a neural network can be adjusted in three different dimensions: depth (more or fewer layers), width (more or fewer channels), or resolution (higher or lower resolution training images). [Reference list]

[0079] 1 Perceptual system

[0080] 2 Optical system

[0081] 3 First optical component

[0082] 4 Second optical component

[0083] 5 Data processing system

[0084] 10 Determining distributions of optical quality with tolerances 11 Calibrating a combination model

[0085] 12 Determining a distribution of the optical quality of the overall optical system

[0086] 13 Determining the functional boundaries of a task model

[0087] 14 Determining test limits

[0088] 30 Device

[0089] 31 storage

[0090] 32 processor

[0091] 33 Entrance

[0092] 34 Exit

[0093] 40 means of transport

[0094] AB shots with windshield

[0095] A T Shots without windshield

[0096] B cBatch of non-degraded images

[0097] B d Batch of degraded images

[0098] B r Batch with residuals

[0099] Bw Batch with recovered images

[0100] BN Bottleneck

[0101] DEC decoder

[0102] ENC Encoder

[0103] F Fourier-optical degradation model

[0104] K Calibrated Confidences

[0105] LT Logit Tensor

[0106] MD1 measurement data for first optical component

[0107] MD2 measurement data for second optical component

[0108] MDG measurement data for the complete optical system

[0109] N Neural Network

[0110] P Calibration block PG Test limit

[0111] RES Restorer

[0112] RN ResNet encoder

[0113] S Segmentation map

[0114] W gMeasured wavefront aberrations W p Predicted wavefront aberrations

Claims

Patent claims 1. Method for determining test limits (PG) for testing the suitability of optical components (3, 4) for use in an overall optical system (2) used by an algorithm, comprising the steps: - Determining (10) tolerance distributions of the optical quality of the individual optical components (3, 4) based on optical measurements in amplitude space; - Determining (12) a distribution of the optical quality of the total optical system (2) by combining the tolerated optical components (3, 4) using a combination model (11) calibrated on the basis of measurements of the optical quality of real total optical systems (2); - Determining (13) functional limits of a task model of the algorithm as a function of the optical quality of the overall system (2); and - Determining (14) test limits (PG) for at least one of the optical components (3, 4) by mapping the functional limits of the task model onto the individual contributions of the optical components (3, 4).

2. Method according to claim 1, wherein the combination model for the tolerated optical components (3, 4) is based on a statistical method.

3. Method according to claim 2, wherein the statistical method uses Latin hypercube sampling or Monte Carlo sampling.

4. Method according to one of the preceding claims, wherein to determine (13) the functional limits of the task model of the algorithm as a function of the optical quality of the total system (2) the optical quality of the total system (2) is used as a perturbation model in the development of the algorithm.

5. Method according to one of the preceding claims, wherein the optical measurements are wavefront measurements.

6. Method according to claim 5, wherein the test limits (PG) for wavefront measurements are determined on the at least one optical component (3, 4) (14).

7. Method according to claim 5 or 6, wherein the measured optical qualities of the optical components (3, 4) and the overall optical system (2) are under The use of a decomposition into an orthonormal functional basis for the description of the measured wavefronts can be parameterized.

8. Method according to claim 7, wherein the measured optical qualities of the optical components (3, 4) and the overall optical system (2) are parameterized by Zernike coefficients or by Legendre coefficients.

9. Method according to one of the preceding claims, wherein the overall optical system (2) is part of a perception system (1).

10. Method according to claim 9, wherein the perception system (1) serves an assistance function or driving function in a means of transport (40).

11. Method according to claim 9 or 10, wherein the optical components (3, 4) are a disk and a camera lens.

12. Method according to one of the preceding claims, wherein the algorithm is implemented as a neural network (N).

13. Method according to claim 12, wherein the neural network (N) is trained for segmentation.

14. Computer program with instructions which, when executed by a computer, cause the computer to perform the steps of a method according to one of claims 1 to 13 for determining test limits (PG) for testing the suitability of optical components (3, 4) for use in an overall optical system (2) used by an algorithm.

15. Device (30) for determining test limits (PG) for testing the suitability of optical components (3, 4) for use in an overall optical system (2) used by an algorithm, comprising a memory (31) in which instructions are stored and a processor (32), wherein the processor (32) is configured to execute the steps of a method according to one of claims 1 to 13 when the instructions are executed.

Images