DE102022101854A1 - Method for analyzing a luminance distribution of light emanating from a lighting device for a vehicle - Google Patents

Abstract

Method for analyzing a luminance distribution of the light emanating from a lighting device for a vehicle, wherein in a method step a digital image (1) is recorded, which corresponds to a luminance distribution of the light emanating from a lighting device for a vehicle, and wherein in a further method step the digital image (1) is changed in such a way that the changed image (5) is a reconstruction of the luminance distribution, which at least essentially only contains the elements or parts of the luminance distribution that can be perceived by a human being.

Description

The present invention relates to a method for analyzing a luminance distribution of the light emanating from a lighting device for a vehicle.

Defective pixels in high-resolution modules are not always visible to the human observer, since they can be outshone by the surrounding pixels. However, if they are visible, they can distract the driver, which can lead to an increased safety risk. Visible pixel errors lead to an unwanted change in the light image and can significantly reduce the subjective impression of the product quality. Predicting whether a defective pixel will lead to a visible change in the light image cannot be predicted purely from the difference in luminance between the defect and the environment and is therefore a more complicated problem than is apparent at first glance.

It would be desirable to have a methodology for predicting the visibility of defective pixels in the light image of a lighting device for a vehicle.

The problem on which the present invention is based is therefore the specification of a method of the type mentioned at the outset, which enables a statement to be made about the perceptibility of elements or parts of a luminance distribution by a person.

According to the invention, this is achieved by a method of the type mentioned at the beginning with the features of claim 1 . The dependent claims relate to preferred embodiments of the invention.

• - ein digitales Bild, das einer Leuchtdichteverteilung des von einer Beleuchtungsvorrichtung eines Fahrzeugs ausgehenden Lichts entspricht, wird aufgenommen,
• - das digitale Bild wird so verändert, dass das veränderte Bild eine Rekonstruktion der Leuchtdichteverteilung ist, in der zumindest im Wesentlichen nur noch die von einem Menschen wahrnehmbaren Elemente oder Teile der Leuchtdichteverteilung enthalten sind.

According to claim 1, the method is characterized by the following method steps:

• - a digital image corresponding to a luminance distribution of the light emanating from a lighting device of a vehicle is recorded,
• - the digital image is changed in such a way that the changed image is a reconstruction of the luminance distribution, which at least essentially only contains the elements or parts of the luminance distribution that can be perceived by a human being.

The changed image can be used to determine, for example, whether existing pixel errors in a high-resolution module of a lighting device lead to a visible change in the light image. Based on this, it can then be decided whether the module of the lighting device can continue to be used despite the pixel errors or whether it has to be replaced by another module.

It can be provided that the digital image is changed using an algorithm that simulates processing steps of the human visual system. Such an approach opens up the possibility of making comparatively reliable statements about the visibility of possible changes in the photograph.

The algorithm can, in particular in a first step, simulate the change in an image due to the optical structure of a human eye. Furthermore, the algorithm can, in particular in a second step, simulate the processing of image information into contrast information, which is carried out in particular by the human brain or by cell types in the retina. In addition, the algorithm can determine a visibility threshold, in particular in a third step. The visibility threshold is used in particular to decide which elements or parts of the luminance distribution can be perceived by a human and which elements or parts of the luminance distribution cannot be perceived by a human, with the elements or parts of the Luminance distribution are not included in the modified image. It can thereby be ensured that the changed image at least essentially only contains the elements or parts of the luminance distribution that can be perceived by a human being. In particular, it is an algorithm based on various perceptual-psychological models that simulate the first processing steps of the human visual system. The algorithm is preferably based on three individual sub-models that are serially linked to one another. The result of the algorithm is a reconstruction of the luminance distribution that only contains the elements that can still be perceived by humans.

It is possible for the changed digital image to be compared with an image generated from a target luminance distribution of the lighting device, in particular to decide whether the light distribution generated by the lighting device corresponds to suitable specifications, taking into account human perception.

It can be provided that the recording of the digital image is a luminance measurement or is converted into a luminance distribution. Both options can ensure that the image to be changed corresponds to a luminance distribution of the light emanating from a lighting device of a vehicle.

There is a possibility that the lighting device is a tail lamp or a headlight, in particular a high-resolution headlight that can generate a light distribution with a large number of pixels. In particular in the case of high-resolution headlights with correspondingly complex modules, it has proven to be very advantageous to have a method for predicting the visibility of defective pixels in the light image using the method according to the invention.

The recorded digital image can be, for example, a recording of a headlight distribution or a recording of a tail light. Corresponding recordings or luminance measurements can serve as input data for the model used. Eye color and age of an observer, for example, and the spatial resolution of the luminance image can be used as additional input parameters.

The method allows the general prediction of the visibility of inhomogeneities in the light image of a lighting device, in particular without adapting the parameters to the specific environment. Provision can likewise be made for the method to enable the position of the inhomogeneity to be determined. One purpose of using the method is therefore to predict the detectability of imperfections, for example in a headlight distribution.

• 1 eine schematische Darstellung eines Beispiels eines erfindungsgemäßen Verfahrens;
• 2 eine beispielhafte Darstellung einer optischen Kontrastempfindlichkeitsfunktion des menschlichen Auges, wobei die Kontrastsensitivität gegen die Ortsfrequenz aufgetragen ist;
• 3 ein beispielhafter Versuchsaufbau zur Aufnahme eines Bildes, das einer Leuchtdichteverteilung des von einem Scheinwerfer ausgehenden Lichts entspricht;
• 4 ein Beispiel eines Bildes, das einer Leuchtdichteverteilung mit Pixelfehlern entspricht;
• 5 das Bild gemäß 4 nach einer Änderung durch ein erfindungsgemäßes Verfahren.

The invention is explained in more detail below with reference to the attached drawings. It shows:

• 1 a schematic representation of an example of a method according to the invention;
• 2 an exemplary representation of an optical contrast sensitivity function of the human eye, the contrast sensitivity being plotted against the spatial frequency;
• 3 an exemplary test setup for recording an image that corresponds to a luminance distribution of the light emanating from a headlight;
• 4 an example of an image corresponding to a luminance distribution with pixel errors;
• 5 the picture according to 4 after modification by a method according to the invention.

Identical and functionally identical parts and areas are provided with the same reference symbols in the figures.

High-resolution headlights enable the generation of lane and symbol projections and thus provide additional information for the driver and other road users. While on the one hand there is the possibility of emphasizing individual pixels, the light pattern should convey an even impression of the luminance distribution without projections and should not show any noticeable differences in intensity between neighboring pixels. Which intensity differences lead to visible gaps in the light pattern depends on a number of parameters.

A method according to the invention can be used to propose a model for predicting the visibility of such intensity gaps for the human observer, a task which is directly linked to contrast detection. Therefore, the model can implement sequentially applied sub-models based on the contrast sensitivity function (CSF).

The detection of objects or spots in a spotlight-lit scene is highly correlated with contrast detection. For street lighting, CIE Report 19/2 [1] proposes using the visibility grade based on psychophysical data measured by Blackwell [2] to predict the detectability of a target in an illuminated environment. The degree of visibility is defined as the ratio between the contrast (between object and background) and the threshold contrast. Although it is a good measure of the detectability of objects under laboratory conditions, but not applicable to the problem solved within the framework of this application, because a spot at an unknown position and not an object is to be recognized.

Blakemore and Campbell [3] postulated that early visual processing mechanisms can be modeled by overlapping channels sensitive to different spatial frequencies. The authors introduced the reciprocal of threshold contrast, called the Contrast Sensitivity Function (CSF). The contrast sensitivity function is measured by showing the viewer images with sine gratings because the frequency of these test patterns can be tuned very precisely.

The contrast sensitivity function is typically presented in a graph of contrast sensitivity versus spatial frequency in cycles per degree (cyc/deg). Sensitivity also changes with luminance. At lower luminances, the sensitivity decreases and the maximum sensitivity shifts to lower frequencies (see 2 ). Since the human visual system is very complex and not fully understood even today, the contrast sensitivity function can only be a simplified model reduced to certain limitations. Nevertheless, it has been used successfully for various applications, such as quality measurements for image compression algorithms [4] or assessing the vision of patients after eye surgery [5].

Joulan et al. [6] are the first to use the contrast sensitivity function in an automotive lighting context to predict the visibility of objects illuminated by a headlight. They propose to apply a multiscale spatial filter to luminance images that simulates the simple contrast perception of human vision. The filter consists of a weighted sum of Differential of Gaussians (DoG). The weights are adjusted such that the resulting filter conforms to the contrast sensitivity function developed by Barten [7]. However, Barten emphasizes that his CSF model is only valid for photopic vision. Night drives, on the other hand, deliver scenarios that are in the mesopic range.

In the example of the method according to the invention, an approach is used which separates the effects of the optical system of the eye (optical contrast sensitivity function) and the neural processing of the retinal image (neural contrast sensitivity function). The optical contrast sensitivity function includes effects such as glare caused by stray light. The neural contrast sensitivity function simulates the receptive fields of the early stages of the human visual cortex. A third part, inducing the threshold contrast sensitivity function, completes the model and allows the visibility of a contrast to be predicted. The selected threshold contrast sensitivity function is valid for mesopic vision.

• - Das Modell ist nur für die foveale Sicht gültig. Dies schließt Winkel von -2°α≤2° ein [8]. Das bedeutet auch, dass Reize nicht in der Peripherie erkannt werden müssen, bevor sie fokussiert werden.
• - Die Kontrastempfindlichkeit ist auch abhängig von der Präsentationszeit des visuellen Reizes [9]. Unterhalb von 4 s [10] Betrachtungszeit ist die Kontrastempfindlichkeit reduziert. Hier werden nur statische Szenarien betrachtet und den Beobachtern wird mehr als 4 s Beobachtungszeit für die Stimuli gegeben. Daher kann die Zeitabhängigkeit der Kontrastempfindlichkeit vernachlässigt werden.
• - Das Modell ist in einem ersten Schritt für achromatische Lichtmuster ausgelegt.
• - Die Beobachtung wird immer bei binokularer Betrachtung des Reizes betrachtet.
• - Bei der Betrachtung von Kontrasten mit einer bestimmten Frequenz für die Dauer von ca. 1 Minute tritt eine Kontrastadaption auf [3], die die Kontrastempfindlichkeit für den adaptierten Frequenzbereich reduziert. Dieser Effekt wird im Modell nicht berücksichtigt. Es wird angenommen, dass der Beobachter nur ca. 30 s Zeit hat, die Reize zu betrachten.
• - Eine besondere Aufmerksamkeit des Beobachters auf bestimmte Regionen führt zu höheren Kontrastempfindlichkeiten [11]. Dieser Effekt wird nicht berücksichtigt.

The described example of a method according to the invention can be a model with a reasonable number of parameters. The following boundary conditions can be taken into account for this:

• - The model is only valid for the foveal view. This includes angles of -2°α≤2° [8]. This also means that stimuli do not have to be detected in the periphery before they are focused.
• - The contrast sensitivity also depends on the presentation time of the visual stimulus [9]. Contrast sensitivity is reduced below 4 s [10] viewing time. Here only static scenarios are considered and the observers are given more than 4 s observation time for the stimuli. Therefore, the time dependence of the contrast sensitivity can be neglected.
• - In a first step, the model is designed for achromatic light patterns.
• - The observation is always viewed with binocular observation of the stimulus.
• - When viewing contrasts with a certain frequency for a period of approx. 1 minute, contrast adaptation occurs [3], which reduces the contrast sensitivity for the adapted frequency range. This effect is not considered in the model. It is assumed that the observer only has about 30 s to look at the stimuli.
• - If the observer pays special attention to certain regions, this leads to higher contrast sensitivities [11]. This effect is not taken into account.

It is entirely possible to choose other boundary conditions for the model.

1. 1. Eine optische Kontrastempfindlichkeitsfunktion 2, die die Abbildungsfehler durch die Optik des menschlichen Auges simuliert. Dazu gehören beispielsweise Effekte wie Blendung, die durch Streuung am Augenmedium erfolgt. Im hier behandelten Beispiel ist das Modell von Watson implementiert [12]. Der Algorithmus simuliert somit in einem ersten Schritt die Veränderung eines Bildes durch den optischen Aufbau eines menschlichen Auges.
2. 2. Eine neuronalen Kontrastempfindlichkeitsfunktion 3, welche die Kontrasterkennungsmechanismen in der menschlichen Netzhaut simuliert. Es handelt sich dabei insbesondere um eine starke Vereinfachung der heute bekannten Prozesse, die im Gehirn ablaufen. Die von Peli [13] entworfene CSF wird im hier behandelten Beispiel für die Anwendung gewählt. Der Algorithmus simuliert somit in einem zweiten Schritt die Verarbeitung einer Bildinformation zu einer Kontrastinformation durch das menschliche Gehirn.
3. 3. Eine Schwellenwert-Kontrastempfindlichkeitsfunktion 4, mit der die Ergebnisse verglichen werden. Dieses Modell dient zur Bestimmung der Schwelle, anhand welcher entschieden wird, welche Elemente noch sichtbar sind und welche Elemente nicht. Die von Wuerger et al. [14] entwickelte Kontrastempfindlichkeitsfunktion wird im hier behandelten Beispiel als am besten geeignet für die gewünschte Anwendung ausgewählt. Sie ist für einen großen Bereich von Leuchtdichten ausgelegt und somit für mesopisches Sehen gültig. Der Algorithmus bestimmt somit in einem dritten Schritt eine Sichtbarkeits-Schwelle.

The model is designed for digital images 1 as input, which correspond to luminance distributions or luminance images. According to the example of a method according to the invention, the digital image 1 is modified with an algorithm that simulates processing steps of the human visual system. The model has the following sub-models or procedural steps, which are applied one after the other (see 1 ):

1. 1. An optical contrast sensitivity function 2 that simulates the aberrations caused by the optics of the human eye. These include, for example, effects such as glare, which is caused by scattering on the eye medium. In the example discussed here, Watson's model is implemented [12]. In a first step, the algorithm thus simulates the change in an image caused by the optical structure of the human eye.
2. 2. A neural contrast sensitivity function 3 that simulates the contrast detection mechanisms in the human retina. In particular, this is a strong simplification of the processes that are known today that take place in the brain. The CSF designed by Peli [13] is chosen for the application in the example discussed here. In a second step, the algorithm thus simulates the processing of image information into contrast information by the human brain.
3. 3. A threshold contrast sensitivity function 4 against which the results are compared. This model is used to determine the threshold that is used to decide which elements are still visible and which elements are not. The Wuerger et al. [14] developed contrast sensitivity function is selected in the example discussed here as the most suitable for the desired application. It is designed for a large range of luminance and is therefore valid for mesopic vision. In a third step, the algorithm thus determines a visibility threshold.

A block diagram of this model is in 1 shown. The digital image 1 is changed in such a way that the changed image 5 is a reconstruction of the luminance distribution, which at least essentially only contains the elements or parts of the luminance distribution that can be perceived by a human being. Each submodel is explained in more detail in the following sections.

The optical contrast sensitivity function 2 describes the aberration of the image due to effects caused by the media of the eye. It can be used to compute the resulting retinal image for a stimulus presented to a human viewer [12]. Including the optical contrast sensitivity function 2 in the model accounts for effects such as glare, which can have a large impact on the overall contrast sensitivity function.

The model implemented here was developed by Watson [12]. The pupil size, the age of the viewer and the eye color influence the aberration and are therefore considered as input parameters for the model.

Watson used data from a large number of wavefront aberration measurements to develop the model. To adapt to the measured aberrations, Zernike polynomials up to the 35th order were used [12]. From the results, he calculated a mean radially symmetric real modulation transfer function and approximated a function that best fits the data. In addition, the effect of the scattered light is included, which reduces the contrasts. Watson uses the Ijspeert et al. Formula found in 1993 [15] to include the influence of scattered light.

mit $r=\sqrt{u^2+v^2},$

u ist die horizontale Ortsfrequenz und v die vertikale Ortsfrequenz. u₁(d) ist ein Polynom-Fit, der die berechnete Kontrastempfindlichkeitsfunktion an unterschiedliche Pupillendurchmesser d anpasst. D berücksichtigt den rein beugungsbegrenzten Teil des Kontrastempfindlichkeitsfunktion. Er wird für weißes Licht mit einer Mittenwellenlänge von λ = 555nm berechnet.As indicated in [12], Watson's model concludes: $$CSf\left(\mathrm{right},\mathrm{i.e}\right)={\left[1+{\left(\frac{\mathrm{right}}{{\mathrm{and}}_1\left(\mathrm{i.e}\right)}\right)}^2\right]}^{-0.62}\cdot \sqrt{D\left(\mathrm{right},\mathrm{i.e},\lambda \right)}\cdot \left(1-S\left(a,p\right)\right)$$

with $\mathrm{right}=\sqrt{{\mathrm{and}}^2+v^2},$

u is the horizontal spatial frequency and v is the vertical spatial frequency. u ₁ (d) is a polynomial fit that fits the calculated contrast sensitivity function to different pupil diameters d. D takes into account the purely diffraction-limited part of the contrast sensitivity function. It is calculated for white light with a center wavelength of λ = 555nm.

S takes into account the effect of stray light. It depends on the pigmentation factor p and the age α of the observer, as well as the given reference age of 70 years. The equations for D, u ₁ (d) and S(a,p) can be found in the publication by Watson ([12], equations (1), (4), (6)).

Finally, as described in Watson et al. [16] calculated. This model is an extension of the model designed by Stanley and Davies [17] and includes aging effects and a distinction between binocular and monocular vision. The model proposed here is considered for binocular vision. This results in the following equation for calculating the pupil diameter: $${\mathrm{i.e}}_{\text{SD}}\left(L,q\right)=7.75-5.75\left(\frac{{\left(\frac{L\cdot q}{846}\right)}^{0.41}}{{\left(\frac{L\cdot q}{846}\right)}^{0.41}+2}\right)$$

L ist die mittlere Leuchtdichte des Sichtfeldes (fov), q die Fläche des fov und a das Alter des Betrachters. Das Referenzalter a_ref ist mit 28,58 Jahren angegeben.The dynamic range of the pupil is reduced with age. This is accounted for in the equation by adding an extra term: $$\mathrm{i.e}\left(L,q,a\right)={\mathrm{i.e}}_{SD}\left(L,q\right)+\left(a-a_{\mathrm{right}ef}\right)\cdot \left(0.02132-0.02132-0.009562\cdot {\mathrm{i.e}}_{SD}\left(L,q\right)\right)$$

L is the mean luminance of the field of view (fov), q is the area of the fov, and a is the age of the viewer. The reference age a _ref is given as 28.58 years.

The implemented neural contrast sensitivity function 3 or neural modulation transfer function was developed by Peli [13]. It is originally applied to natural images or complex scenes and allows the prediction of whether small details in the image are visible to the human observer.

The image is convolved with various cosine-log bandpass filters, each with different center frequencies and a bandwidth of one octave. The filters in the spatial frequency domain are calculated to $$G_k\left(\mathrm{right}\right)=0.5\cdot \left(1+\text{cos}\left(\mathrm{\pi }\cdot {\text{<figure-callout id="2" label="log" filenames="DE102022101854A1\_0011.png" state="{{state}}">log</figure-callout>}}_2\text{right}-\mathrm{\pi }\cdot k\right)\right)$$

Each filter has a center frequency of 2 ^k , where k is an integer value.

The filters designed here are very similar to the Gabor filters, with the difference that the sum of the filters is equal to one [13]. Hubel and Wiesel [18] find that receptive fields are very similar to Gabor filter functions. The model is therefore a simplified approach to simulating early stages of visual processing because it includes effects such as orientational selectivity as suggested by de Valois et al. [19] was discovered is not taken into account.

The image is convolved with each filter separately, giving k results. $$a_k\left(x,y\right)=ƒ\left(x,y\right)\ast G_k\left(x,y\right)$$

where f(x,y) is the image value at horizontal pixel position x and vertical pixel position y, gk(x,y) is the _kth bandpass filter function in the spatial domain, and * represents the convolution operator. For the model exemplified, enlarges the image by half the filter size in each direction before filtering. The values at the edges of the image are repeated to avoid artificial edging. The resolution information of the image is not changed by this process and after filtering the image size is reduced back to the original size. This so-called padding avoids ringing artefacts caused by the discrete Fourier transformation and is a common method when multiplying filters in the frequency domain [20].

Since the model explained as an example is designed for foveal vision, the field of vision is limited to ± 2 degrees [8]. The area required to represent a full cycle at the lowest frequency should not exceed this range. Adhering to the Peli model design rule that the center frequency is 2 ⁿ , the smallest frequency ƒ _c1 = 2 ^-2 cyc/deg is selected. The maximum applicable frequency for the construction used is limited by the Nyquist frequency. Since the luminance images shown here have a resolution of ~ 75 pix/deg, the maximum frequency that can be achieved by Fourier transformation is 35 pix/deg. This results in a maximum center frequency of ƒ _c8 = 32 cyc/deg. The resulting frequencies, which were explicitly chosen for the application of the model proposed here, are thus given as: 0.5, 1, 2, 4, 8, 16 and 32 cyc/deg.

For photopic lighting environments, a healthy and young human observer is known to have a maximum resolution of 0.5' [21]. This corresponds to a resolution of 120 cycles/degree. In contrast, the highest resolution of the model is set to only a quarter of this value.

2 shows the contrast sensitivity calculated with the model explained as an example for different luminance levels. If one compares the maximum contrast sensitivity at 200 cd/m ² with the contrast sensitivity of 0.2 cd/m ² , the maximum sensitivity drops to around a quarter of the value. Thus, the selected resolution should be sufficient for the model.

To account for the neural adaptation processes, Peli calculates the contrast for each channel by dividing the filtered images by an adaptation luminance value. The value is calculated by keeping only the frequencies in the image that are below the passband of the bandpass filter. The bandpass filtered image is then divided by the result: $$c_k\left(x,y\right)=\frac{a_k\left(x,y\right)}{ƒ\left(x,y\right)\ast l_k\left(x,y\right)}$$

where l _k (x,y) is the low-pass filter in the spatial domain that is convolved with the image.

The calculated contrast is compared to the threshold contrast for each pixel in each contrast image. If the contrast for the given bandpass center frequency is less than a given threshold, the information in the bandpass filtered image at that pixel is discarded by setting the pixel to a value of zero. $$a_k\left(x,y\right)\{\begin{array}{ll}ƒ\left(x,y\right)\ast G\left(x,y\right),\hfill & iƒc_k\left(x,y\right)\ge c_{\text{tresh}}\hfill \\ \mathrm{0,}\hfill & else\hfill \end{array}$$

Peli uses the measured CSF of each individual observer as a threshold [22]. After processing each bandpass filtered image in this way, the resulting image is reconstructed by a summation of all filtered and thresholded images, including the low and high frequency residuals. $${ƒ}_{\text{rec}}\left(x,y\right)=l_0\left(x,y\right)+{\displaystyle \sum _{\text{k}=1}^{\text{n}-1}{a}_k\left(x,y\right)+H_n\left(x,y\right)}$$

where l ₀ is the low-pass residual and h _n is the high-pass residual.

Although Peli's results agree well with his study [23], this method is not directly applicable to the model proposed here. The goal is to develop a general approach, not one tailored to individuals. For this reason, a model that predicts a mean contrast sensitivity function for healthy subjects is used to calculate the threshold.

The environment in which headlamps are typically used has luminance levels below 3 cd/m ² , where the transition between photopic and mesopic vision takes place [24]. It is therefore important to choose a contrast sensitivity function that is valid for the range of mesopic vision. The contrast sensitivity function selected as the threshold function for this model was developed for adaptation luminances between 0.02 cd/m ² and 7000 cd/m ² [14]. It also contains a separable part describing the chromatic contrast sensitivity function. This allows the possible extension of the model in a later step without having to implement another threshold function.

Wuerger et al. develop the model as a continuous function dependent on the surrounding luminance and frequency, allowing calculation for any average luminance or frequency found in the measured luminance distribution. They validate the model with their own measurement data and with an appropriate comparison to other published data sets. The authors propose an extension of the model that includes the dependence of the contrast sensitivity function on the number of cycles represented. This extension is not applied here because the authors state that the data set to verify the extension is not large enough. So the resulting achromatic logarithmic contrast sensitivity function is calculated as follows: $$S_{\text{log10}}\left(\mathrm{right},{\mathrm{right}}_{\text{Max}},b\right)={\text{log}}_{10}\left(s_{\text{Max}}\right)-{\left(\frac{{\text{log}}_{10}\left(\mathrm{right}\right)-{\text{log}}_{10}\left({\mathrm{right}}_{\text{Max}}\right)}{0.5\cdot 2^b}\right)}^2$$

where r is the radial frequency as used in the previous sections, r _max is the frequency at which the contrast sensitivity function has its maximum value, and S _max is the maximum sensitivity at the value r _max . The calculations of the values for r _max and S _max can be found in Wuerger et al, [14], equations (16a,b).

The threshold contrast used for comparison can then be calculated by c _thresh = 10 ^{S log10} .

The luminance L used to calculate the threshold is the mean luminance that is also used to calculate the contrast in (6).

In order to test the applicability of the model explained by way of example to headlight light patterns, it is necessary to generate a large variety of light patterns. 3 shows an example test setup for recording an image that corresponds to a luminance distribution of the light emanating from a headlight.

A camera 7 is mounted 2 m behind a projector 6 and 1.2 m above a road 8. The center of the projector lens is 0.64 m above the road. Camera 7 and projector 6 are thus placed in positions very similar to the position of the headlights and the driver. In particular, the projector 6 can project a light distribution 9 onto the road 8, which can correspond to that of a headlight.

For exemplary tests, a high-performance Barco W30 Flex projector is used as projector 6, which is geometrically and luminous intensity calibrated (because it receives 8-bit gray scale values as input data). The maximum luminous flux of the projector 6 is specified as 30,000 lumens. With an image size of 1920 x 1200 pixels and a corresponding field of view (fov) of ±15.03° in the horizontal direction and ±10.05° in the vertical direction, the resolution of the projector 6 is 0.017° vertically and 0.016° horizontally. The projector 6 is located in a light channel that enables a stable test environment that is independent of the time of day and the weather.

The model is tested with measured luminance images. A luminance measuring camera of the type LMK5 Color, Techno Team, is used as camera 7, which produces the luminance images using a lens with a focal length of 16 mm and a neutral density filter with a transmission of 7.93%.

A grid of points is used to translate the pixel positions into angular coordinates. The center of each point is spaced 0.5° apart in the vertical and horizontal directions. An image processing algorithm is used to measure the center point of the points in the luminance image in pixel coordinates. In a next step, the position is linked to the corresponding angle. Bilinear interpolation between the measured points gives each pixel in the image an angular coordinate in degrees. After the angular positions of the projector have been determined in this way, the angular positions for the camera image can be calculated by knowing the distance to the center of the projector.

The light pattern used for the test is generated using a simulated light intensity distribution of a headlight module, which consists of a high-resolution, pixelated light source. The simulation enables individual pixels to be dimmed and switched off completely.

1. 1. Es wird ein dunkler Fleck auf einem gleichmäßig beleuchteten Hintergrund erzeugt, indem ausgewählte Pixel abgeschaltet werden. Die Größe des Flecks wird dann durch Abschalten benachbarter Pixel verändert. Es wird erwartet, dass sich die Sichtbarkeit mit der Größe des dunklen Flecks verbessert.
2. 2. Ein heller Fleck wird auf dem gleichen Hintergrund erzeugt und die Größe des Flecks wird auf die gleiche Weise wie zuvor geändert. Es wird das gleiche prinzipielle Verhalten wie für den dunklen Fleck erwartet.
3. 3. Bei dem letzten Szenario wird eine komplexe Szene einer typischen Landstraße mit stationärer Straßenbeleuchtung erzeugt. Die Auswirkung der lokalen Adaption auf die Auflösung wird getestet. Es wird erwartet, dass die Sehschärfe in dunkleren Bereichen abnimmt.

Three different scenarios are examined:

1. 1. A dark spot is created on an evenly lit background by turning off selected pixels. The size of the spot is then changed by turning off adjacent pixels. Visibility is expected to improve with the size of the dark spot.
2. 2. A bright spot is created on the same background and the spot is resized in the same way as before. The same principle behavior as for the dark spot is expected.
3. 3. In the last scenario, a complex scene of a typical country road with stationary streetlights is created. The impact of local adaptation on resolution is tested. Visual acuity is expected to decrease in darker areas.

The first two scenarios are initially based on an in 3 projected white screen, not shown, which is 8.4 m away from the projector 6. Then the screen is removed and the same image is projected onto the street 8. For the third scenario, the light pattern is projected onto the road 8 only. For the scenarios shown, the observer's age is set to 30 years and the observer's eye color is assumed to be brown.

The first scenario examines dark spots that change in size. Spot sizes with a diameter of 0.2°, 0.3° and 0.4° are generated by dimming neighboring pixels of the light source to different extents. The selected simulation results in an ambient light intensity of 16770 cd and a spot light intensity of 11690 cd. The schematic representation in 4 shows an example of some small dark spots 10 and two large dark spots 11.

The luminance images recorded by the camera 7 are filtered using bandpass filters with different center frequencies. Bandpass filters with small center frequencies respond to spatially larger elements in the image and vice versa. In particular, bandpass filters with 1, 2 and 4 cyc/deg respond the most to the image.

Contrasts and threshold contrasts are calculated for the luminance images filtered in this way. Because the contrast threshold changes with frequency, different thresholds are calculated for each center frequency. A comparison of the contrast thresholds on the road with the contrast threshold on the screen shows that the contrast thresholds change with the luminance. Darker regions have a significantly higher threshold value.

The smaller the spot size, the greater the blur between the background and the screen. While some of this effect is due to the reduced contrast caused by blurring of neighboring pixels, the model still shows the effect because it is larger in the reconstruction than in the original.

An exemplary image reconstructed using a method according to the invention for projection onto the road is shown in 5 evident. It turns out that the visibility is reduced for smaller spot sizes until they are no longer or hardly visible. In particular, the in 4 pictured small dark spots 10 in 5 no longer visible, whereas the large spots 11 can still be clearly recognized. These results make it clear that the model can be used for the desired area of application.

In the second scenario, bright spots that change in size are examined. For the assessment of the recognition quality of bright spots on an evenly lit background, the same ambient light level as in the first scenario is used for the simulated light source. The pixels that create the spots are set to a lower dimming level (higher intensity) than the surrounding area. The light intensity for the bright spot is given as 21490 cd.

The investigation makes it clear that the bright spots appear blurrier and less visible in the reconstructed images compared to the original image. This corresponds to the observable fact that a bright spot was more difficult to distinguish from its surroundings than a dark spot of the same size.

In the third scenario, a complex scene is examined, in particular a more complex scene of a typical urban street at night. A low-beam distribution illuminates the road in the presence of stationary streetlights.

It can be seen that for regions in the image with low luminance, a significantly lower resolution of the reconstructed image is observed than for regions with higher luminance. Again, this is consistent with the behavior of the human visual system. In a dark environment, the visual acuity of the visual system is reduced.

The qualitative behavior of the model thus agrees with the expected behavior for all three scenarios. This is a very good indicator of the applicability of the model. The advantage of the model is its general applicability for a variety of luminance distributions and environments without the need for parameter adjustment. By using the three partial models, effects such as physiological glare and global as well as local luminance adaptation effects are taken into account in the contrast calculation.

bibliography

1. [1] Commission Internationale de l'Eclairage, "An Analytic Model for Describing the Influence of Lighting Parameters upon Visual Performance", CIE publication 19/2, Vienna, 1981.
2. [2] H.R. Blackwell, "Contrast thresholds of the humane eye", Journal of the Optical Society of America, 1946.
3. [3] Blakemore C, Campbell FW, On the existence of neurones in the human visual system selectively sensitive to the orientation and size of retinal images, Journal of Physiology, 203 (1), pp. 237-260, 1969
4. [4] MJ Nadenau, J. Reichel, and M. Kunt, "Wavelet-based color image compression: exploiting the contrast sensitivity function.", IEEE Transactions on image processing, 12.1, pp: 58-70, 2003
5. [5] Yamane N, Miyata K, Samejima T, Hiraoka T, Kiuchi T, Okamoto F, Hirohara Y, Mihashi T, and Oshika T Ocular higher-order aberrations and contrast sensitivity after conventional laser in situ keratomileusis”, Investigative Ophthalmology & Visual Science, 45(11), pp.3986-3990, 2004.
6. [6] K Joulan, N Hautiere, and N Bremond, “Contrast sensitivity function for road visibility estimation on digital images”, Proc. 27th session of the Commission Internationale de l'Eclairage, 2011
7. [7] P.G.J. Barten, "Contrast sensitivity of the human eye and its effects on image quality", SPIE press, 1999
8. [8] S.E. Palmer, Vision Science. Photons to Phenomenology, MIT Press, Cambridge, Massachusetts, 1999.
9. [9] Van Nes FL, Koenderink JJ, Nas H, and Bouman MA, "Spatiotemporal Modulation Transfer in the Human Eye," Journal of the Optical Society of America A, 57, 1082-1088, 1967
10. [10] B. Hauser, H. Ochsner, and E. Zrenner. "The "Blendvisus"-Part 1: Physiological basics of the change in vision with increasing test field luminance." Clinical monthly sheets for ophthalmology (200.02), pp. 105-109, 1992
11. [11] N.V.K. Medathati, H. Neumann, GS. Masson, and P. Kornprobst, "Bio-inspired computer vision: Towards a synergistic approach of artificial and biological vision", Computer Vision and Image Understanding, vol. 150, pp. 1-30, 2016.
12. [12] A.B. Watson, "A formula for the mean human optical modulation transfer function as a function of pupil size." Journal of Vision, 13.6, 18-18, 2013
13. [13] E Peli, Contrast in complex images. Journal of the Optical Society of America A, 7, 10, 2032-2040, 1990
14. [14] Wuerger S, Ashraf M, Kim M, Martinovic J, Pérez-Ortiz M, and Mantiuk RK, "Spatio-chromatic contrast sensitivity under mesopic and photopic light levels", Journal of Vision, 20(4 ), 2020
15. [15] JK Ijspeert, TJTP Van Den Berg, and H Spekreijse, "An improved mathematical description of the foveal visual point spread function with parameters for age, pupil size and pigmentation." Vision research 33(1) pp: 15-20 , 1993.
16. [16] AB Watson, and JI Yellott, "A unified formula for light-adapted pupil size.", Journal of vision, 12.10, 12-12, 2012
17. [17] P.A. Stanley, and A.K. Davies, "The effect of field of view size on steady-state diameter pupil.", Ophthalmic & Physiological Optics, 15(6), pp. 601-603, 1995
18. [18] I.E. Hubel, and T. Wiesel, "Receptive fields, binocular interaction, and functional architecture in the cat's visual cortex", Journal of Physiology, London, 160:106-154, 1962
19. [19] R.L. de Valois, DG Albrecht, and L.G. Thorell, "Spatial frequency selectivity of cells in macaque visual cortex", Vision Research, 22(5), pp.545-559,1982
20. [20] A Distante, and C Distante, Handbook of Image Processing and Computer Vision. Volume 1: From Energy to Image, p. 438, Springer International Publishing, 2020.
21. [21] J.A. Ferwerda, "Elements of early vision for computer graphics.", IEEE computer graphics and applications 21.5 pp: 22-33, 2001
22. [22] E Peli, "Test of a model of foveal vision by using simulations.", Journal of the Optical Society of America A, 13.6, 1131-1138, 1996
23. [23] E Peli, "Contrast sensitivity function and image discrimination," Journal of the Optical Society of America A, 18, 283-293, 2001
24. [24] B. Wördenweber, P. Boyce, D.D. Hoffmann and J. Wallaschek, "Automotive lighting and human vision", Vol. 1., Springer-Verlag, Berlin Heidelberg, 2007.

Reference List

1

digital image

2

optical contrast sensitivity function

3

neural contrast sensitivity function

4

Threshold contrast sensitivity function

5

changed picture

6

projector

7

camera

8th

road

9

light distribution

10

small dark spot

11

big dark spot

Claims (10)

Method for analyzing a luminance distribution of the light emanating from a lighting device for a vehicle, characterized by the following method steps: - a digital image (1), which corresponds to a luminance distribution of the light emanating from a lighting device for a vehicle, is recorded, - the digital image (1 ) is changed in such a way that the changed image (5) is a reconstruction of the luminance distribution, which at least essentially only contains the elements or parts of the luminance distribution that can be perceived by a human being. procedure after claim 1 , characterized in that the digital image (1) is modified with an algorithm that simulates processing steps of the human visual system. procedure after claim 2 , characterized in that the algorithm, in particular in a first step, simulates the change in an image (1) through the optical structure of a human eye. Procedure according to one of claims 2 or 3 , characterized in that the algorithm, in particular in a second step, simulates the processing of image information into contrast information, which is carried out in particular by the human brain or by cell types in the retina. Procedure according to one of claims 2 until 4 , characterized in that the algorithm, in particular in a third step, determines a visibility threshold. procedure after claim 5 , characterized in that the visibility threshold serves to decide which elements or parts of the luminance distribution are perceivable by a human and which elements or parts of the luminance distribution are imperceptible by a human, the according to this decision not perceptible by a human Elements or parts of the luminance distribution are not included in the modified image (5). Procedure according to one of Claims 1 until 6 , characterized in that the changed digital image (5) is compared with an image generated from a target luminance distribution of the lighting device, in particular to decide whether the light distribution generated by the lighting device, taking into account human perception, corresponds to suitable specifications. Procedure according to one of Claims 1 until 7 , characterized in that the recording of the digital image (1) is a luminance measurement or is converted into a luminance distribution. Procedure according to one of Claims 1 until 8th , characterized in that the lighting device is a tail light or a headlight, in particular a high-resolution headlight that can generate a light distribution (9) with a large number of pixels. Procedure according to one of Claims 1 until 9 , characterized in that the recorded digital image (1) is a recording of a headlight distribution or a recording of a tail light.

Images