3.1.1.

Optical layout

Figure 2(a) shows the optical layout of a CWB sensor that features three spectral channels similar to traditional RGB color cameras. First, incoming light passes through a telescope formed by an external camera lens and a collimator C ($f=28\mathrm{mm}$ and $f/\#=2.0$). Subsequently, the light is split by two successive dichroic beam splitters ${\mathrm{DBS}}_1$ and ${\mathrm{DBS}}_2$ into three channels corresponding to blue light ($\sim 400$ to 500 nm), green light ($\sim 500$ to 600 nm), and red light ($\sim 600$ to 700 nm). The spectral separation of the three bands is improved by the use of an appropriate band-pass filter ${\mathrm{BP}}_x$ in each channel. This is necessary to attenuate out-of-band laser radiation effectively. Finally, the light is focused on the respective (monochrome) imaging sensor ${\mathrm{S}}_x$ by lens ${\mathrm{L}}_x$ ($f=25\mathrm{mm}$ and $f/\#=1.4$). A photograph of our laboratory demonstrator, built with standard optical and optomechanical components, is shown in Figs. 2(b) and 2(c). Table 1 lists the optical elements used to implement the three-channel CWB sensor.

Fig. 2

(a) Scheme of the optical layout of the three-channel CWB sensor. (b), (c) Photographs of the laboratory demonstrator (without external camera lens).

Table 1

Optical elements used to implement the three-channel CWB sensor.

The telescope (camera lens and collimator C) at the start of the optical path is not necessary to implement the CWB sensor. However, it offers an intermediate focal plane that can be used for various purposes. The original intention was to introduce a nonlinear optical limiter at this position in order to protect the imaging sensors additionally against laser damage. Furthermore, a calibration target can be placed there, for example, to verify the correct alignment of the imaging sensors.

Please note that the band-pass filters depicted by single elements in the scheme of Fig. 2(a) maybe realized in the laboratory demonstrator by a combination of a long-pass and a short-pass filter (see Table 1). Furthermore, the short-pass filter Edmund Optics 84714 used to form band-pass filter ${\mathrm{BP}}_3$ is located in front of the collimator C in our laboratory demonstrator. Thus, it acts as an IR cutoff filter for the whole system.

3.1.2.

Transmittance

To estimate the sensor’s performance, we calculated the spectral transmittance of the three channels. For that purpose, we used transmittance data of the optical elements listed in Table 1, as specified by the manufacturer; an external camera lens was not taken into account. For data not available in digital form, we digitized transmittance curves provided by the manufacturer in graphical form. Figure 3 shows the calculated transmittance for all three channels in diabatic scale.

Fig. 3

Calculated spectral transmittance of the three-channel CWB sensor (without external camera lens).

The calculations were performed both for unpolarized and s-/p-polarized light since the transmittance of the dichroic beamsplitters (DBSs) is polarization-dependent. The different colors (blue, green, and red) in the graphs correspond to the different channels, whereas the line styles of the curves (solid, dashed, and dotted) show the polarization state of the light (unpolarized, s-polarized, and p-polarized). Additionally, colored bands in the graphs highlight the range of values between minimum and maximum transmittance.

From the calculations, we expected out-of-band values of the transmittance ranging from ${10}^{-6}$ to ${10}^{-8}$ [corresponding to an optical density (OD) ranging from 6 to 8] for each channel. This would be a reasonable value to protect the imaging sensors against out-of-band laser radiation and, thus, keeping their image information in case of a laser attack. To verify the calculations, we performed measurements, which are presented in Sec. 4.1.

3.1.3.

Image fusion

It may be advantageous if the CWB sensor’s output is only a single image that is shown to the operator instead of presenting all three images acquired by the three individual imaging sensors. Therefore, the three channel’s monochrome images are fused together. For the image fusion process, there is a multitude of possibilities. The simplest approach would be to calculate the mean of all three images, which would result in a monochrome image. Since the design of the three-channel CWB sensors offers three spectral channels corresponding roughly to blue, green, and red light, it seems obvious to generate a colored fused image just by putting the three images into the channels of an RGB image. Both methods are illustrated in Fig. 4.

Fig. 4

Result of different image fusion methods applied to the three images (B, G, R) of the three-channel CWB sensor. $M$, calculation of the mean; $MF$, calculation of the mean with filtering of overexposed image areas; $RGB$, creation of an RGB image; $RGBF$, creation of an RGB image with filtering of overexposed image areas.

The upper row of Fig. 4 shows the three input images, as acquired by the three-channel CWB sensor. The labels $B$, $G$, and $R$ correspond to channel 1, channel 2, and channel 3, respectively (see Fig. 2). In this example, the CWB sensor is dazzled by laser radiation with a wavelength of 656 nm; the image of channel 3 is nearly completely overexposed. The output image resulting from the simple calculation of the mean is labeled $M$; the color fused image is labeled $RGB$. We can see that in the case of laser dazzle, the partial or complete failure of one channel leads to a decrease in image contrast for both methods, although the structural image information is still there. Therefore, we decided to implement some kind of filtering prior to image fusion.

The simplest approach is to analyze the three channels for overexposed/saturated pixels. For filtering, we apply the rule that only saturated pixels (of corresponding image areas) occurring not simultaneously in all three channels shall be filtered out by the fusion algorithm. We define a pixel as saturated when its pixel value is larger than the (arbitrarily chosen) threshold value of 250 for 8-bit images (maximum pixel value is 255). In that case, the pixel values of the overexposed areas are neglected for the calculation of the mean image or the generation of the color image. For the case that in all three channels, corresponding image areas are overexposed, we attribute this to a (natural) broadband light source and the filtering process does not take place. The result of the filtering process is shown in Fig. 4 by the images labeled $MF$ and $RGBF$ for the calculation of the mean image and the generation of the color image, respectively.

We learn that this kind of filtering leads to a monochrome image with good contrast since, in this example, the fused image is mainly composed of the sensor images $B$ and $G$. In the case of the color image RGBF, the filtering process also improves the contrast. However, there is a strong color distortion because the signal of channel 3 ($R$) is missing. Therefore, we usually operate the three-channel CWB sensor with fusion method MF. Since this is unsatisfactory in some way, we thought about a way to implement the principle of CWBs in a sensor system, which is able to deliver color output images without color distortion in the case of laser dazzle. We were able to realize this in a two-channel sensor system based on multiband optical elements. This two-channel CWB sensor is presented in the following Sec. 3.2.

3.2.1.

Optical layout

We improved the optical layout of our CWB concept by utilizing multiband optical elements (band-pass filters and DBS), which are characterized by alternating high and low transmittance windows. Based on a multiband beamsplitter and two appropriate multiband band-pass filters, we set up a two-channel CWB sensor. The optical layout is shown in Fig. 5(a); a photograph of the two-channel CWB sensor with the external camera lens is given in Fig. 5(b). The optical elements for its realization are listed in Table 2.

Fig. 5

(a) Scheme of the optical layout of a two-channel CWB sensor and (b) photograph of the laboratory demonstrator.

Table 2

Optical elements used to implement the two-channel CWB sensor.

Similar to the three-channel CWB sensor, the incoming light passes through a telescope formed by an external camera lens and a collimator C ($f=25\mathrm{mm}$ and $f/\#=1.4$). Then, the light is split by a multiband DBS into two channels. The spectral separation of the two bands is increased by the use of appropriate multiband band-pass filters ${\mathrm{BP}}_1$ and ${\mathrm{BP}}_2$. The spectral transmittance of the multiband optical elements is shown in Fig. 6. The red lines correspond to the DBS; the blue and green lines correspond to the band-pass filters for channel 1 and channel 2, respectively. The transmittance is plotted for unpolarized light (solid line), s-polarized light (dashed line), and p-polarized (dotted line) light.

Fig. 6

Polarization-dependent transmittance of the multiband optical elements used to implement the two-channel CWB sensor.

3.2.2.

Transmittance

Based on the transmittance curves of the optical elements shown in Fig. 6, we estimated the transmittance of the two channels. As for the three-channel CWB sensor, we used the data provided by the manufacturers or digitized graphical data if necessary. Figure 7 shows a plot of the transmittance as a function of wavelength for channel 1 and channel 2 in diabatic scale.

Fig. 7

Calculated spectral transmittance of the two-channel CWB sensor (internal optics only).

The two colors (blue and green) in the graphs distinguish the different channels. As before, the line styles of the curves (solid, dashed, and dotted) show the polarization state of the light (unpolarized, s-polarized, and p-polarized). Colored bands in the graphs highlight the range of values between minimum and maximum transmittance. For the two-channel CWB sensor, we expect out-of-band transmittance values of ${10}^{-8}$ (corresponding to an OD of $\sim 8$) for each channel.

From the graph of Fig. 7, we can recognize the advantage of using optical multiband elements. Both channels receive light from the blue, green, and red parts of the visible spectrum. Therefore, we can use color imaging sensors for the two channels, which allows us to receive two independent color images of the same scene section. If one channel is dazzled by laser radiation, the complementary channel will still provide an undisturbed color image.

3.2.3.

Image fusion

For the two-channel CWB sensor, the image fusion process is the same as for the three-channel CWB sensor, except that RGB color images are processed instead of monochrome images. This is illustrated in Fig. 8, where the sensor system is illuminated with laser radiation of wavelength 532 nm.

Fig. 8

Result of different image fusion methods applied to the two images (Ch. 1, Ch. 2) of the two-channel CWB sensor. $M$, calculation of the mean; $MF$, calculation of the mean with filtering of overexposed image areas.

A mean image is calculated from the two acquired input images (labeled Ch. 1 and Ch. 2 in Fig. 8). The calculation of the mean image can be performed by taking the two images just as they are. This results in an output image that may have reduced contrast when one channel is dazzled by laser light (labeled $M$ in Fig. 8). Alternatively, as explained for the three-channel CWB sensor, we can introduce filtering of overexposed image areas in advance of the mean image calculation. This leads to an output image with high contrast (labeled $MF$ in Fig. 8).

4.2.1.

Definition

We now introduce the quantity “spectral separation” $S{S}_{xy}$ of two individual channels $x$ and $y$ at a specific wavelength. We define this quantity as the absolute value of the difference of the OD values ${\mathrm{OD}}_x$ and ${\mathrm{OD}}_y$ for channel $x$ and $y$ multiplied by a factor of 10:

By using the factor 10, we will state values of spectral separation using the unit decibel (dB).

The spectral separation tells us how well these channels are spectrally separated at a specific wavelength. Let us assume that one of our CWB sensors is affected by laser radiation and channel $x$ shall be dazzled. If at the concerning laser wavelength, the spectral separation $S{S}_{xy}$ of channels $x/y$ is quite large, it means that the value of ${\mathrm{OD}}_y$ of channel $y$ is very different at this wavelength. In case ${\mathrm{OD}}_y$ is larger than ${\mathrm{OD}}_x$, the vulnerability of channel $y$ to laser dazzle would be lower compared with channel $x$. As long as the laser power is not too high to also dazzle channel $y$, the CWB sensor can provide image information to the user.

Contrarily, a very low value of the spectral separation $S{S}_{xy}$ means that the OD values of channels $x$ and $y$ are very similar. Thus, laser radiation will also have a similar effect on both channels. This does not necessarily mean that the threatening laser would dazzle both channels since the OD values of both channels could be similar but very large.

Thus, the value of the spectral separation alone is not a sufficient measure of the CWB sensor’s resilience against laser dazzle. However, it gives a hint at which wavelengths the CWB sensor is hardened against laser dazzle and at which wavelengths problems may occur. The reason why we introduce this quantity is that it can also be easily measured for other devices, e.g., color cameras using an imaging sensor with Bayer filter mosaic, whereas the transmittance/OD may not be measurable in a simple way.

For example, in case of a sensor with Bayer filter mosaic, we cannot measure the transmittance of the different filters directly since we do not have access to the space behind the filters. Also, in the case of three-sensor cameras, the transmittance of the channels can only be measured, if the camera is disassembled, which involves the danger of damaging the camera or, at least, misaligning the three imaging sensors. For such cameras, however, it is possible to measure the spectral separation using the signal of the imaging sensor(s) when irradiated with light, as will be described in the following.

The digital signal DS of a pixel of an imaging sensor can be calculated by³³

Now, we assume that the dark signal ${\mathrm{DS}}_{\text{dark}}$ of an imaging sensor is negligible or that the acquired images have been corrected accordingly (dark-frame correction). Furthermore, the irradiance of an imaging sensor’s channel $x$ of a CWB sensor is proportional to its channel transmittance: $E_x=T_x{E}_{\mathrm{in}}$. The digital signal of channel $x$ can then be written as

Since the different imaging sensors of a multisensor camera (three-sensor camera or CWB sensor) are usually of the same type (i.e., same quantum efficiency, etc.), the ratio of transmittances for channel $x$ and $y$ is given by

Equation (4) can be used together with Eq. (1) to calculate the spectral separation of two channels of a camera sensor, when values of transmittance/OD cannot be measured directly. In Eq. (4), we have assumed that the exposure times $t_{\exp ,x}$ and $t_{\exp ,y}$ of the channels’s imaging sensors can be different to be able to measure higher ratios of $T_x/T_y$.

4.2.2.

Measurements

We measured the spectral separation of three COTS camera devices (see Fig. 12) in order to compare it with our CWB sensors. The first one was a standard color CMOS camera using an imaging sensor with Bayer filter mosaic (Allied Vision Mako G-158C), the second one was a three-CMOS camera (JAI AP-1600T-PGE), and the third one was a hyperspectral imager using a $4\times 4$ filter mosaic (Photonfocus MV1-D2048x1088-HS03-96-G2).

Fig. 12

COTS camera devices: (a) Color CMOS camera Allied Vision Mako G-158C, (b) three-CMOS camera JAI AP-1600T-PGE, and (c) hyperspectral imager Photonfocus MV1-D2048x1088-HS03-96-G2.

Some technical data regarding the imaging sensors of these cameras as well as of our CWB sensors are listed in Table 3. The standard CMOS camera and the three-CMOS camera have been selected in such a way to have the same imaging sensor.

Table 3

Parameters of the imaging sensors utilized in the cameras under test.

Our experimental setup to measure the spectral separation is shown in Fig. 13. Since we expected rather low values of spectral separation ($\le 50\mathrm{dB}$) for the three COTS camera devices, we used an incoherent, broadband halogen light source (Thorlabs SLS201/M). A set of 27 band-pass filters BP (Thorlabs FBxxx-10) was used to generate narrowband light, with (nominal) center wavelengths ranging from 470 to 730 nm in steps of 10 nm. The (nominal) FWHM of the band-pass filters was 10 nm. Behind the band-pass filter, the light was coupled into a bifurcated fiber bundle (Thorlabs BFY400LS02, one input connector and two output connectors) using a fiber coupler ${\mathrm{FC}}_1$. The first fiber output was connected to a power meter for monitoring, and the second fiber output was connected to a fiber collimator ${\mathrm{FC}}_2$ (Thorlabs RC08SMA-P01). The collimated output beam was sent to an aperture (6 mm diameter) and subsequently directed to the camera sensor. For the measurements, all camera sensors were equipped with the same camera lens (Kowa LM25NC3, $f=25\mathrm{mm}$).

Fig. 13

Experimental setup to measure the spectral separation of the spectral channels of a camera sensor.

The procedure of the spectral separation measurement was as follows: We first chose a wavelength by inserting the respective band-pass filter. Then, the camera’s exposure time $t_{\exp }$ was chosen so to produce a strong signal but no saturation. For later analysis (to estimate the digital signal DS), an image was acquired using this exposure time. In the course of the experiments, the exposure time was increased stepwise in order to generate also a signal in another color channel. This procedure was repeated until all color channels delivered a signal at all individual measurement wavelengths.

As an example, Fig. 14 shows three camera images acquired with the three-CMOS camera at the wavelength of 510 nm. Figures 14(a)–14(c) correspond to exposure times of 30, 500, and $\mathrm{40,000}\mu \mathrm{s}$, respectively. Since the camera lens was adjusted to infinity, it had the effect that the end facet of the fiber was imaged onto the imaging sensor. Thus, the camera images showed a centrally located disk with quite homogeneous pixel values (see the line profiles in Fig. 14).

Fig. 14

Camera images acquired with the three-CMOS camera JAI AP-1600T-PGE using the experimental setup shown in Fig. 13 ($\lambda =510\mathrm{nm}$) for different exposure times of (a) $30\mu \mathrm{s}$, (b) $500\mu \mathrm{s}$, and (c) $\mathrm{40,000}\mu \mathrm{s}$. Additionally, the pixel values along the horizontal lines marked in the images by red dashed lines are shown in the graphs below.

For the analysis, we extracted the pixel values of a rectangular part located within the central disk (see the orange squares in the images of Fig. 14) and calculated the mean values for each of the $n$ color channels for that area. Using the mean values ${\mathrm{DS}}_x$ and the corresponding exposure times $t_{\exp ,x}$ noted during the measurement ($x$ ranging from 1 to $n$), we calculated the mutual spectral separation of the channels using Eqs. (1) and (4).

For a camera system with $n$ spectral channels, the number of mutual ratios of channel transmittance would be $n(n-1)/2$. This results in one spectral separation value for the two-channel CWB sensor and in three values ($\text{red}\leftrightarrow \text{green}$, $\text{red}\leftrightarrow \text{blue}$, and $\text{green}\leftrightarrow \text{blue}$) for a three-channel RGB systems like the standard color camera, the three-CMOS camera, or our three-channel CWB sensor. The hyperspectral imager exhibits 16 passbands, which yields 120 mutual values of spectral separation.

4.2.3.

Results

We measured the spectral separation of three COTS devices using the aforementioned experimental setup and procedure. For our CWB sensors, we used the data of the transmittance measurements (see Sec. 4.1) to calculate the spectral separation. For comparison and in contrast to the transmittance measurements described in Sec. 4.1, we also performed a measurement on our two-channel CWB sensor in the same way as for the COTS devices. But with the difference that we used an unpolarized, tunable supercontinuum light source (NKT SuperK Compact + NKT SuperK Varia) instead of the incoherent light source, as described in the experimental setup of Fig. 13. The reason the halogen light source could not be applied was that the spectral purity of these band-pass filters is just not good enough for cases with very high spectral separation larger than $\sim 50\mathrm{dB}$. In Fig. 15, the spectral separation as a function of wavelength is plotted for all devices tested.

Fig. 15

Separation of the spectral channels of different cameras as a function of wavelength: (a) Color CMOS camera Allied Vision Mako G-158C using a single imaging sensor with Bayer filter mosaic. (b) Three-CMOS camera JAI AP-1600T-PGE, (c) hyperspectral imager Photonfocus MV1-D2048x1088-HS03-96-G2 using a $4\times 4$ filter mosaic (16 passbands), (d) three-channel CWB sensor, and (e) two-channel CWB sensor.

The graphs of Figs. 15(a)–15(c) show the results for the COTS devices; the graphs of Figs. 15(d) and 15(e) are the results for our three-channel and two-channel CWB sensors. Typically, all plots include the mutual spectral separation of the individual channels. Regarding the results presented in Fig. 15(c), however, we decided to plot only the maximum values of the 120 curves of spectral separation to avoid overfilling the graph. The maximum values are also shown in Figs. 15(a) and 15(b) for the single-sensor camera and the three-CMOS camera, respectively. For the plots regarding our CWB sensors, we also included the calculated values as hatched bands just as for the plots of transmittance of Figs. 10 and 11.

The interpretation of the plots has to be done carefully. In principle, a high value of spectral separation is desirable. Thus, the maximum curve shown in the graphs of Figs. 15(a)–15(c) is the most interesting curve regarding the assessment of a sensor’s vulnerability to laser dazzle. If a sensor system has a high (maximum) value of spectral separation at a specific wavelength, it means that at least one of its spectral channels has a high value of OD at this wavelength and will (to a certain degree) be saved from laser dazzle. When we look at the graphs of Figs. 15(a)–15(c), we can see that the spectral separation ranges from 30 to 50 dB for the three-CMOS camera and is below 20 dB for the standard color camera and the hyperspectral imager. This means that the three-CMOS camera will be less vulnerable to information loss due to laser dazzle than the other two cameras. Values around 50 dB are quite good; we did not expect such high values for the three-CMOS camera before the measurements.

Looking at the spectral separation of our two CWB sensors, we can see that the value of spectral separation is quite high (60 to 80 dB) within the channel’s passbands. At the crossover points of the channel’s transmittance curves (see Figs. 10 and 11), the spectral separation drops to low values. However, this does not necessarily mean that the CWB sensors are vulnerable to laser dazzle at these wavelengths. For example, the spectral separation of the three-channel CWB sensor drops nearly to zero at a wavelength of around 500 nm. Looking at the transmittance plots of Fig. 10, we can see that the transmittance values of all spectral channels are very similar for these wavelengths ($\mathrm{OD}\sim 7$ to 8), resulting in a low spectral separation. However, due to the high OD values and very low transmittance, the sensor is not vulnerable to laser dazzle at wavelengths around 500 nm. Altogether, we can conclude that our CWB sensors are much less vulnerable to laser dazzle than COTS camera devices.

4.3.1.

Measurements

For this work, we performed laser dazzle experiments with our CWB sensors as well as with the COTS devices presented in Sec. 4.2. Additionally, we performed the laser dazzle experiments with a standard monochrome CMOS camera, Allied Vision Mako G-158B, for purposes of comparison. This monochrome camera is identical to the color camera Allied Vision Mako G-158C introduced in Sec. 4.2, except that the imaging sensor is not equipped with the Bayer filter mosaic.

A sketch of the experimental setup is shown in Fig. 16(a). The sensor under test observed a white screen at a distance of 5.14 m. A test pattern was projected on the screen using a video projector (Optoma UHD60). As test pattern, we decided to use a highly structured, fractal test pattern according to the work of Landeau³⁷ [see Fig. 16(b)]. Such a test pattern consists of a large number of black and white squares arranged in a particular manner; details can be found in the publication of Landeau. For our measurements, we prepared the fractal test pattern in such a way that a single square in the fractal test pattern had an angular size of 0.1 deg as seen by the sensor. Sensor dazzling was performed using a multiwavelength laser source iChrome MLE from Toptica offering four different laser wavelengths (488, 515, 561, and 640 nm). A hole in the screen’s center allowed laser illumination of the sensor along its optical axis. At the position of the sensor, the laser beam diameter ($1/{\mathrm{e}}^2$) was $\sim 16\mathrm{cm}$. Maximum irradiance at the position of the sensor was $\sim 638$, 279, 606, and $401\mu \mathrm{W}/\mathrm{c}{\mathrm{m}}^2$ using a laser wavelength of 488, 515, 561, and 640 nm, respectively.

Fig. 16

(a) Experimental setup for the assessment of laser protection measures. (b) Image section of the fractal test pattern projected onto the screen for the measurements.

For the measurements, we set the sensor’s exposure time to a value such that the white areas of the test pattern deliver a digital signal of half of the sensor’s dynamical range, e.g., a digital signal of 127 for an 8-bit sensor with a maximum pixel value of 255. Accomplishing this condition was not always possible since the video projector is based on digital light processing technology. The projector uses a single digital micromirror and a rotating color wheel to produce color images. When the sensor’s exposure time is not tuned to the rotary frequency of the filter wheel, it can happen that color changes occur in successive sensor images. For our video projector, the rotary frequency of the filter wheel seems to be 120 Hz, because the sensor’s exposure time had to be set to a multiple of $8.333\mathrm{ms}=1/120\mathrm{Hz}$ in order to acquire images without color distortion. Therefore, it was not possible to equal the exposure conditions, the mean pixel value of white areas ranged from 105 to 165 for the different camera devices.

The dazzle experiments were set up in such a way that each sensor was illuminated with a series of different levels of irradiance while images were taken each time. This procedure was carried out for all laser wavelengths (488, 515, 561, and 640 nm). Subsequently, we analyzed the images by calculating the SSIM index for all images. Figure 17 presents some examples of images taken with the different camera sensors at a laser wavelength of 640 nm. The camera images shown were taken at two extreme situations: for a rather low value of irradiance of $\sim 10\mu \mathrm{W}/\mathrm{c}{\mathrm{m}}^2$ (measured at the entrance of the camera lens) and for the maximum irradiance of $\sim 400\mu \mathrm{W}/\mathrm{c}{\mathrm{m}}^2$. Since the sensors have slightly different fields of view (FOVs), we additionally indicated matching image areas by green rectangles drawn into the camera images. The indicated image areas were used for further analysis, as will be described hereafter.

Fig. 17

Sensor images taken according to the experimental setup of Fig. 16: (a) Color CMOS camera Allied Vision Mako G 158C, (b) monochrome CMOS camera Allied Vision Mako G-158B, (c) three-CMOS camera JAI AP-1600T-PGE, (d) hyperspectral imager Photonfocus MV1-D2048x1088-HS03-96-G2, (e) three-channel CWB sensor, and (f) two-channel CWB sensor.

4.3.2.

Data analysis

To compare the SSIM results of our different camera sensors, we have to take into account that the sensors have different FOVs and different image resolutions. The calculated value of SSIM will depend on these parameters since changes in FOV and resolution cause differences in the image information. Therefore, in order to be able to compare the SSIM results of the different sensors, we had to match the FOV and the resolution of the image data before the SSIM calculations were made.

The images of all sensors had to be cropped in a first step to match the favored FOV. The matching image areas are indicated in the example images of Fig. 17 by green rectangles. In a second step, the image data were sampled down to equalize the final image resolution. In this step, the image resolution was defined by the cropped image of the three-channel sensor, which had the lowest resolution. The final size of the processed images was $581\times 365\text{pixels}$. Using the processed image data, the SSIM calculations were performed. For our CWB sensors, we analyzed the fused image, which is monochrome in case of the three-channel CWB sensor and colored in case of the two-channel CWB sensor (although the test chart is monochrome).

4.3.3.

Results

As an example, Fig. 18 presents detailed results gained for the color CMOS camera Allied Vision Mako G-158C. It shows four plots of the SSIM index as a function of irradiance corresponding to the four different laser wavelengths. For the color channels, the SSIM index was calculated separately (curves labeled $R$, $G$, and $B$ in the plots). Additionally, the mean SSIM value for all color channels was calculated [curve labeled $Mean(R,G,B)$ in the plots]. Furthermore, the color image was converted to a monochrome image by taking the mean of the color channels before calculating the SSIM index (curve labelled $RGB\to Mono$ in the plots). The latter two curves are quite similar but not equal. In the plots, a colored background highlights the area between these two curves.

Fig. 18

Results of the SSIM analysis of the color CMOS camera Allied Vision Mako G-158C. (a) 488 nm, (b) 515 nm, (c) 561 nm, and (d) 640 nm.

Looking at the plots of Fig. 18, we recognize a behavior as expected. The SSIM curve for the color channel corresponding to the laser wavelength starts to decrease earlier with increasing irradiances as compared with the other channels. The out-of-band channels are less affected.

Figure 19 shows an overview of the results of the SSIM analysis for all sensors under test. Since the amount of information given by Fig. 18 is overwhelming, we just kept the maximum SSIM curve for each wavelength in the plots of Fig. 19. We can see that both of our CWB sensors perform best in keeping the image information under laser dazzle followed by the three-CMOS camera. The results for the three-CMOS camera are quite good, especially for the red laser wavelength of 640 nm, since the spectral separation of the three-CMOS camera peaks at this wavelength (see Fig. 15). The hyperspectral imager performs better than the single-sensor color and monochrome cameras except for the laser wavelength of 640 nm.

Fig. 19

Results of the SSIM analysis for (a) the color CMOS camera Allied Vision Mako G-158C, (b) the monochrome CMOS camera Allied Vision Mako G-158B, (c) the three-CMOS camera JAI AP-1600T-PGE, (d) the hyperspectral imager Photonfocus MV1-D2048x1088-HS03-96-G2, (e) the three-channel CWB sensor, and (f) the two-channel CWB sensor.