A fast detection method of light wave polarization direction based on image grayscale signature

Abstract

The present invention discloses a method for rapid detection of polarization direction of light waves based on local grayscale signature of an image, including: S1, offline operation, for generating a circumferential pixel coordinate table, a radial pixel coordinate table, a grayscale weight table and a template grayscale signature; S2, online operation, for making a rough estimate of the darkest direction according to the circumferential pixel coordinate table, obtaining a polarization image grayscale signature according to the rough estimate of the darkest direction, the radial pixel coordinate table and the grayscale weight table, and correlating the polarization image grayscale signature with the template grayscale signature to obtain an accurate estimate of the polarization direction. The present invention has the advantages of both high precision and high speed, and can fully meet the requirements of real-time detection of polarization direction based on vector polarized light field.

Description

Technical Field

[0001] The present invention relates to the field of polarization detection technology, and more specifically, to a method for quickly detecting the polarization direction of light waves based on image grayscale signatures. Background Art

[0002] Polarization characteristics are an important carrier of optical information. The detection of polarization direction plays an important role in polarization remote sensing reconnaissance, thin film parameter measurement, polarization imaging, optical fiber communication monitoring and other fields. At present, there are four types of commonly used polarization measurement systems: interference type, spectroscopic type, time modulation type and spatial modulation type. Among them, spatial modulation polarization detection technology has significant advantages such as high system stability, simple optical path structure, and single measurement results. However, traditional spatial modulation polarization detection methods mostly use spatial polarization modulation devices (such as spatial polarizers, polarization gratings, gradient refractive index lenses, and birefringent prisms) to achieve spatial modulation of the light waves to be measured. Since these spatial modulators are difficult to manufacture, the further development of spatial modulation polarization detection technology is limited. In order to break through this limitation, researchers proposed a polarization detection technology based on spatial modulation of vector polarization light fields. Taking advantage of the fact that the polarization distribution of the vector polarized light field has a regular spatial variation, the linearly polarized light or partially polarized light can be modulated by means of the first-order vector angular polarized light field (radial polarized light field), so as to form a polarization image with a wedge-shaped light and dark distribution of light intensity, and the direction of the diameter of the darkest pixel (the darkest direction angle) is parallel (perpendicular) to the vibration direction of the incident polarized light vector. Therefore, the digital image processing technology can be used to process the intensity image of the polarized light collected by the camera, and the darkest direction angle of the wedge can be detected to achieve the measurement of the polarization direction of the incident light. The basic principles and schemes of this type of technology have been effectively verified by multiple research groups. The Chinese invention patent (201510641478.7) "Sky polarization pattern detection method and system based on s-wave plate" published by Zhang Wenjing et al. uses global Radon transform to extract the polarization direction angle contained in the image. This method has the problems of being sensitive to noise and large amount of calculation. In the Chinese invention patent (201810331028.1) "An image processing method for calculating the polarization direction and polarization degree of light waves", Lei Bing et al. used a method related to the local angle of the image to extract the central angle of the image region of interest, which improved the detection accuracy. However, it still has the defects of being time-consuming and unable to meet the needs of real-time tasks.

[0003] In view of the above analysis, it is necessary to study a method for rapid detection of light wave polarization direction based on image grayscale signature, aiming to quickly and accurately extract the polarization direction angle from the polarization image, laying the foundation for the practical application of polarization detection technology based on vector polarized light. Summary of the invention

[0004] The purpose of the present invention is to provide a method for quickly detecting the polarization direction of light waves based on image grayscale signature, so as to overcome the defects of low accuracy and long time consumption in the prior art.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] The method for quickly detecting the polarization direction of light waves based on image grayscale signature includes the following steps:

[0007] S1, offline operation, used to generate a circumferential pixel coordinate table, a radial pixel coordinate table, a grayscale weight table and a template grayscale signature;

[0008] S2, online operation, is used to make a rough estimate of the darkest direction based on the circular pixel coordinate table, quickly obtain the grayscale signature of the polarization image based on the rough estimate of the darkest direction, the radial pixel coordinate table and the grayscale weight table, and correlate the grayscale signature of the polarization image with the grayscale signature of the template to obtain an accurate estimate of the polarization direction.

[0009] Furthermore, the step of generating the circumferential pixel coordinate table in step S1 is:

[0010] For an M×N image, a polar coordinate system is established with the center pixel of the image [round(M / 2), round(N / 2)] as the origin, and the polar coordinates (r p ,θ p );

[0011] Set a set of sampling radii {R1, R2, ..., R n}, for each sampling radius R i (i=1,2,...,n), if the pixel point p satisfies R i -r ξ ≤r p ≤R i +r ξ , then the pixel coordinates (u p ,v p ) is stored in radius R i The corresponding circle pixel coordinate table.

[0012] Among them, round(ξ) means rounding the data ξ to the nearest integer, r ξ It is the radius error range when searching for circular pixels.

[0013] Furthermore, the steps of generating the radial pixel coordinate table and the grayscale weight table in step S1 are:

[0014] The polar coordinates (r p ,θ p) is used as the basis, and the image with a radius range of [R min ,R max ] every pixel point p(u p ,v p ) is divided into 4 sub-pixels, whose pixel coordinates are (u p -0.25,v p -0.25)、 (u p -0.25,v p +0.25)、(u p +0.25,v p -0.25) and (u p +0.25,v p +0.25);

[0015] In the angle range [0°, 180°), the angle θ st For each sampling angle Draw a straight line through the origin Calculate each sub-pixel p s (u ps ,v ps ) to the straight line Vertical distance

[0016] If any sub-pixel p s (u ps ,v ps ) to the straight line The vertical distance is less than the set threshold d th , then it is considered that the pixel p(u p ,v p ) is included in the above straight line, and the pixel coordinates (u p ,v p ) recorded in In the corresponding radial pixel table, the weight w of the pixel point is p Recorded in The corresponding grayscale weight table.

[0017] Among them, (x ps ,y ps ) is the sub-pixel p s (u ps ,v ps ), u0 and v0 represent the matrix row and column coordinates of the image center p0, respectively. [ξ] represents the largest integer not greater than ξ, belongs to pixel p and to the straight line The distance is less than the threshold d th The sum of all sub-pixel weights of

[0018] Furthermore, the method for generating the template grayscale signature in step S1 comprises the following steps:

[0019] With the pixel point with pixel coordinates of [round(M / 2),round(N / 2)] as the center, generate an ideal polarization image with the darkest direction of 0° and size of M×N as a template;

[0020] The angle coverage and sampling interval of the grayscale signature feature are set to ±θ ri With θ si , with 0° as the center, for the i-th sampling angle (% represents the remainder operation), query the radial pixel coordinate table and the grayscale weight table to obtain the radial grayscale weighted integral value in, Indicates the sampling angle The corresponding radial pixel coordinate table, I p (u j ,v j ), w p (u j ,v j ) are pixels p(u j ,v j ) and the gray value of the pixel at angle θ si The corresponding weight value in the grayscale weight table;

[0021] The grayscale radial integral values ​​calculated at each sampling angle are arranged in counterclockwise order and normalized to eliminate the influence of ambient brightness to obtain the template grayscale signature.

[0022] Furthermore, in the method for generating the template grayscale signature feature, if the influence of noise is taken into consideration: the deviation angle range is set to ±θ rc With interval θ sc , for the i-th deviation angle by The grayscale signature feature of the template is calculated as the center, and % represents the remainder operation.

[0023] Furthermore, the step of roughly estimating the darkest direction according to the circumferential pixel coordinate table in step S2 is:

[0024] For the polarization image I0 to be measured, find the pixel coordinates ([u0, v0]) of the center p0 of the region where the grayscale is wedge-shaped distributed in the image I0;

[0025] For each sampling radius R set in the offline operation in step S1 i(i=1,2,...,n), query the circular pixel coordinate table corresponding to each sampling radius, and according to each coordinate [u ri ,v ri ], extract pixel coordinates from image I0 as [u0,v0]+[u ri ,v ri ]-[round(M / 2),round(N / 2)] pixel grayscale values, and record the grayscale values ​​of the pixels in the circle that are less than the threshold value T g The pixel polar angle;

[0026] Calculate the average value θ of the polar angles of the pixels extracted from each circle min The sampling angle θ is set when the radial pixel coordinate table is generated in the offline operation in step S1. st is the precision versus the mean value θ min Round off to get a rough estimate of the darkest direction angle.

[0027] Furthermore, the step of obtaining the grayscale signature of the polarization image to be measured according to the rough estimated value of the darkest direction, the radial pixel coordinate table and the grayscale weight table in step S2 specifically includes:

[0028] Centered on the roughly estimated darkest direction angle, the angle range ±θ covered by the grayscale signature ri and the sampling interval θ si , query the radial pixel coordinate table and grayscale weight table corresponding to each sampling angle;

[0029] According to the data obtained from the radial pixel coordinate table [u i ,v i ], extract the pixel coordinates of the image to be measured I0 as [u0,v0]+[u i ,v i ]-[round(M / 2),round(N / 2)] pixel grayscale, and combined with the pixel weight obtained from the grayscale weight table, calculate the grayscale signature of the polarization image to be measured.

[0030] Wherein, [u0, v0] is the pixel coordinate of the center p0 of the grayscale wedge-shaped distribution area in the polarization image I0 to be measured, M and N are the number of rows and columns of the image matrix, respectively, and round() represents the rounding operation of the data in the brackets.

[0031] Furthermore, the step of correlating the polarization image grayscale signature with the template grayscale signature to obtain an accurate estimation of the polarization direction in step S2 specifically includes:

[0032] The image grayscale signature g is respectively compared with the template grayscale signature corresponding to each deviation angle obtained in the offline operation Perform correlation and search for the deviation angle corresponding to the template grayscale signature with the maximum value θ rc With θ sc The deviation angle range and interval set by the algorithm are respectively used to obtain an accurate estimate of the polarization direction. If the first-order vector angular polarized light field is used for modulation, the darkest direction angle is the polarization direction of the polarized light; if the first-order vector radial polarized light field is used for modulation, the polarization direction of the incident light is perpendicular to the darkest direction, that is, the polarization direction is

[0033] arg(ξ) returns the parameter that satisfies the condition ξ, θ min is a rough estimate of the darkest direction angle.

[0034] Compared with the prior art, the advantages of the present invention are:

[0035] 1. The present invention can effectively improve the algorithm's anti-noise ability, has good robustness, and greatly reduces the requirements for image quality: after extracting the image grayscale signature features from the polarization image, the precise polarization direction angle will be determined by the correlation operation results between the features. The correlation operation is a highly robust operation. The introduction of this operation greatly suppresses the interference of noise, thereby achieving accurate polarization direction extraction.

[0036] 2. The present invention greatly improves the efficiency of the algorithm and is very suitable for real-time measurement tasks. The specific reasons are:

[0037] ① The main body of the method avoids global search and adopts a strategy that combines rough estimation based on circumferential data with precise search in a small angle range;

[0038] ② In the correlation operation adopted in the small angle range search, the large amount of image pixel information is abandoned and the grayscale signature feature with smaller amount of data is adopted;

[0039] ③The accuracy of this method is mainly determined by the deviation angle interval set when the template feature is generated. Therefore, the extraction of grayscale signature features allows for a larger angle sampling interval;

[0040] ④ If the algorithm parameters remain unchanged, different polarization image processing processes can share the circular pixel table, radial pixel table, grayscale weight table, and template grayscale signatures at different deviation angles. Therefore, the above tables are generated by the offline operation stage, and the online operation only needs to use the table lookup method to calculate the grayscale signature features of the image to be measured. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is a flow chart of a method for rapid detection of polarization direction of light waves based on image grayscale signature of the present invention;

[0043] Figure 2 It is a display of multiple groups of circular pixels generated offline;

[0044] Figure 3 This is a schematic diagram of grayscale signature generation;

[0045] Figure 4 is the grayscale signature feature of the template image and one of the deviation angles;

[0046] FIG5 is a polarization image to be measured and a set of circumferential data therein in an embodiment of the present invention;

[0047] Figure 6 is the grayscale signature feature of the image to be measured in the implementation case of the present invention;

[0048] Figure 7 is the correlation result between the image feature to be measured and the template feature at each deviation angle in the implementation case of the present invention;

[0049] Figure 8 is a display of the polarization direction detected in the embodiment of the present invention in the image;

[0050] Fig. 9 The graphs are result graphs of measuring polarization images of different polarization directions using three methods: the method provided by the present invention (CE+GS), the traditional global radon method (GR), and the small angle range image correlation method based on rough estimation of the region of interest detection (IAD+LC).

[0051] Fig.10 is one of the real images to be measured in the implementation case of the present invention;

[0052] Fig.11 These are the real image measurement results obtained using three different methods. DETAILED DESCRIPTION

[0053] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0054] Based on the use of first-order vector polarized light modulation to generate a polarization image with a wedge-shaped intensity distribution, the present invention first roughly estimates the darkest direction angle based on multi-circular data, then extracts the grayscale signature features of the image, and correlates them with the grayscale signature of the template, thereby achieving fast and accurate detection of the polarization direction of light waves.

[0055] In the specific implementation, after the optical system is built, the size and quality of the polarization images obtained are consistent. During the processing, different images can share algorithm parameters and related data. Therefore, in order to reduce the redundancy of the algorithm and improve the real-time operation speed, the solution proposed by the present invention includes two parts: offline operation and online operation. The circumferential pixel table, radial pixel coordinate table, grayscale weight table and template grayscale signature features are generated through offline operation. The online operation obtains the grayscale signature of the polarization image and obtains the precise polarization direction through feature correlation. The specific contents are as follows:

[0056] See also Figure 1 As shown, this embodiment discloses a method for quickly detecting the polarization direction of light waves based on image grayscale signature, comprising the following steps:

[0057] Step S1, offline operation, is used to generate a circumferential pixel coordinate table, a radial pixel coordinate table, a grayscale weight table and a template grayscale signature.

[0058] The steps of generating the circular pixel coordinate table are as follows: for an M×N image, a polar coordinate system is established with the center pixel point [round(M / 2), round(N / 2)] of the image as the origin, and the polar coordinates (r p ,θ p ); Set a set of sampling radii {R1, R2, ..., R n}, for each sampling radius R i (i=1,2,...,n), if the pixel point p satisfies R i -r ξ ≤r p ≤R i +r ξ , then the pixel coordinates (u p ,v p ) is stored in radius R i The corresponding circular pixel coordinate table; round(ξ) means rounding the data ξ to an integer, r ξ is the radius error range when searching for circular pixel points, and M and N are the number of rows and columns of the polarization image acquired by the system, respectively.

[0059] The steps of generating the radial pixel coordinate table and the grayscale weight table are as follows:

[0060] The polar coordinates (r p ,θ p ) is used as the basis, and the image with a radius range of [R min ,R max ] every pixel point p(u p ,v p ) is divided into 4 sub-pixels, whose pixel coordinates are (u p -0.25,v p -0.25)、 (u p -0.25,v p +0.25)、(u p +0.25,v p -0.25) and (u p +0.25,v p +0.25); in the angle range [0,π), at an angle θ st For each sampling angle Draw a straight line through the origin Calculate each sub-pixel p s (u ps ,v ps ) to the straight line The vertical distance d ps :

[0061]

[0062] (x ps ,y ps ) is the sub-pixel p s (u ps ,v ps )’s image coordinates:

[0063]

[0064] u0 and v0 represent the matrix row and column coordinates of the image center p0, respectively. s (u ps ,v ps ) to the straight line The vertical distance is less than the set threshold d th , then it is considered that the pixel p(u p ,v p ) is included in the above line, and the pixel coordinates (u p ,v p ) recorded in In the corresponding radial pixel table, the weight w of this point is p Recorded in The corresponding grayscale weight table. Among them:

[0065]

[0066] [ξ] represents the largest integer not greater than ξ, and w p belongs to pixel p and to the straight line The distance is less than the threshold d th The sum of all sub-pixel weights, that is:

[0067]

[0068] is the i-th sub-pixel of pixel p to The vertical distance.

[0069] The method for generating the template grayscale signature includes the following steps: According to Malus's law, the grayscale value of each pixel in the ideal polarization image with the darkest direction of 0° has cos 2 (θ t +θ p -π / 2) change law (θ p is the pixel polar angle). With the pixel point with pixel coordinates of [round(M / 2), round(N / 2)] as the center, generate an ideal polarization image with the darkest direction of 0° and size of M×N as the template; set the angle coverage range and sampling interval of the grayscale signature feature to ±θ respectively. ri With θ si , with the darkest direction as the center, for the i-th sampling angle Query the radial pixel coordinate table and grayscale weight table to obtain the radial grayscale weighted integral value:

[0070]

[0071] % represents the remainder operation, Indicates the sampling angle The corresponding radial pixel coordinate table, I p (u j ,v j ), w p (u j ,v j ) are pixels p(u j ,v j ) and the gray value of the pixel at angle θ si The corresponding weight value in the grayscale weight table. Each sampling angle The calculated grayscale radial integral values ​​are arranged in counterclockwise order and normalized to eliminate the influence of ambient brightness, thus obtaining the grayscale signature of the template.

[0072] Theoretically, when the algorithm parameters remain unchanged, for images with different polarization directions, the grayscale signature features obtained with the darkest direction as the center will be completely consistent. However, due to the influence of noise, there is a certain deviation between the rough estimate of the darkest direction of the polarization image to be measured and the true value during the online operation stage. For this reason, the deviation angle range ±θ is set. rc With interval θ sc , for the i-th deviation angle With this angle as the center, the grayscale signature of the template is calculated, thereby obtaining and storing the grayscale signatures of the template under multiple deviation angles. % represents the remainder operation.

[0073] Step S2, online operation, is used to make a rough estimate of the darkest direction by querying the circular pixel coordinate table, and obtain the grayscale signature of the polarization image based on the rough estimate of the darkest direction, the radial pixel coordinate table and the grayscale weight table, and correlate the grayscale signature of the polarization image with the grayscale signature of the template to obtain an accurate estimate of the polarization direction.

[0074] The steps for roughly estimating the darkest direction by querying the circumferential pixel coordinate table are as follows:

[0075] For the polarization image I0 to be measured, find the pixel coordinates ([u0, v0]) of the center p0 of the region where the grayscale is wedge-shaped distributed in the image; for each sampling radius R set in the offline operation in step S1 i (i=1,2,...,n), query the circular pixel coordinate table corresponding to the sampling radius, and according to each coordinate [u ri ,v ri ], extract pixel coordinates from image I0 as [u0,v0]+[u ri ,v ri ]-[round(M / 2),round(N / 2)], and take the center of the region p0 as the origin to calculate the grayscale value of the pixel less than the threshold T g The polar angle average of all pixels is then calculated using the sampling angle θ set when generating the radial pixel coordinate table in the offline operation in step S1. st The average is rounded for accuracy to give a rough estimate of the darkest direction angle Where [u0, v0] is the pixel coordinate of the center p0 of the grayscale wedge-shaped distribution area in the polarization image I0 to be measured, M and N are the number of rows and columns of the image matrix, and round() represents the rounding operation of the data in the brackets. It should be noted that since the value range of the polarization direction is [0, 180°), the polar angle of [-180°, 0) needs to be transformed into the range of [0, 180°) during the calculation process.

[0076] The step of obtaining the grayscale signature of the polarization image to be measured according to the rough estimated value of the darkest direction, the radial pixel coordinate table and the grayscale weight table specifically includes: obtaining the darkest direction angle obtained by the rough estimation Centered on the grayscale signature, the angle range covered is ±θ ri and the sampling interval θ si , query the radial pixel coordinate table and grayscale weight table corresponding to each sampling angle; according to the data obtained from the radial pixel coordinate table [u i ,v i ], extract the pixel coordinates of the image to be measured I0 as [u0,v0]+[u i ,v i ]-[round(M / 2),round(N / 2)], and combined with the pixel weights obtained from the grayscale weight table, the grayscale signature of the polarization image to be measured is calculated. [u0,v0] is the pixel coordinate of the center p0 of the grayscale wedge-shaped distribution area in the polarization image I0 to be processed, M and N are the number of rows and columns of the image matrix respectively, and round(ξ) represents the rounding operation of the data.

[0077] The step of correlating the polarization image grayscale signature with the template grayscale signature to obtain an accurate estimate of the polarization direction specifically includes:

[0078] The grayscale signature g of the image to be measured is respectively compared with the template grayscale signature corresponding to each deviation angle obtained in the offline operation Perform correlation and search for the deviation angle corresponding to the template grayscale signature with the maximum value

[0079]

[0080] θ rc With θ sc are the deviation angle range and interval set by the algorithm respectively, and arg(ξ) is the parameter operation that returns the condition ξ.

[0081] So far, we can deduce that:

[0082]

[0083] θ min A rough estimate of the angle for the darkest direction obtained during online operation.

[0084] If the first-order vector angular polarized light field is used during modulation, the darkest direction angle of the polarized image, that is, the polarization direction of the polarized light, can be obtained through formula (7); if the first-order vector radial polarized light field is used during modulation, the darkest direction angle of the polarization direction of the incident light is vertical, that is, the polarization direction is:

[0085]

[0086] The following is a detailed description of the embodiments of the present invention in conjunction with the accompanying drawings and measured data:

[0087] In the example of the present invention, the size of the image to be processed is 1601×1601. Therefore, in the offline operation, the pixel point with the pixel coordinates of [801,801] is taken as the center of the circle, the sampling radius is in the range of 300 to 450 pixels, the sampling interval is 5 pixels, the radius error is 1 pixel, and 16 groups of circular pixels are generated. Figure 2 Marked in blue. Figure 2 In this example, the template image is generated by computer with a size of 1601×1601 and a darkest direction of 0°.

[0088] In the implementation case of the present invention, the horizontal right direction is set to 0°, and the angles of the other directions are the counterclockwise rotation angles from 0° to that direction. In offline operation, within the range of [0,180°), with an angle interval of 0.01°, a total of 18,000 radial pixel coordinate tables and grayscale weight tables corresponding to different angles are generated. The deviation angle search range is set to ±0.3°, and the search interval is 0.01°. By querying the radial pixel coordinate table and grayscale weight table generated in the offline operation, combined with formula (5), the template grayscale signatures corresponding to a total of 61 deviation angles {-0.3°, -0.29°, ..., 0, ..., 0.29, 0.3°} are obtained respectively. When calculating the grayscale signature, the angle coverage range and sampling interval of the grayscale signature feature are set to ±45° and 0.6°, respectively. Figure 3 An example of angle sampling during grayscale signature calculation corresponding to a 179.7° deviation angle is shown. Figure 4 The template grayscale signature features corresponding to the deviation angle are displayed, containing a total of 151 data.

[0089] FIG5(a) is a polarization image to be measured in an example of the present invention, and its image pixel size is 1601×1601. The center pixel coordinates of the region with a wedge-shaped grayscale distribution in the image are (790, 801), which is marked by a red 'x' in the figure. The inner and outer radii of the circular area are set to 300 pixels and 450 pixels respectively, with a radius sampling interval of 5 pixels and a radius error of 1 pixel. For each radius value within [300, 310, 320, ..., 450], the circular pixel coordinate table corresponding to the radius generated in the offline operation is queried, and the coordinate data [u i ,v i ], extract pixel coordinates as [u i -11,v i]. Figure 5(b) shows the distribution of the grayscale of pixels on a circle with a radius of 360 pixels (the red circle shown in Figure 5(a)) with polar angles. To show the grayscale of each pixel, the polar angle range in this figure is [0°, 360°). Due to the influence of noise, there are many burrs on the grayscale modulation curve, but it can still be found that it has cos 2 (θ t +θ p -π / 2) change law (θ p is the polar angle, θ t is the darkest angle). Through online operation, Figure 4 The rough angle estimate based on multi-circle data is 115.051°. Since the angle interval set in the offline operation is 0.01° in the example of the present invention, the angle average is rounded to this accuracy, and the final rough estimated angle value of the darkest direction is 115.05°.

[0090] According to the angle coverage range (±45°) and sampling interval (0.6°) of the grayscale signature feature set in the example of the present invention, for each angle within [75.05, 75.06, ..., 160.05], the radial pixel coordinate table and grayscale weight table corresponding to the angle are queried, and the grayscale weight table is calculated according to the data [u i ,v i ], extract pixel coordinates as [u i -11,v i ], and combined with formula (5), we can get Figure 4 The grayscale signature of the image is shown in Figure 6 .

[0091] The image grayscale signature obtained in the example of the present invention is correlated with the template grayscale signatures at 61 different deviation angles obtained in the offline operation. The correlation results are as follows: Figure 7 As shown. Analysis of the relevant data shows that the image data has the highest matching degree with the template data corresponding to the deviation angle of 0.05 degrees. Therefore, combining formula (6) and formula (7), the darkest direction angle is 115°. Since the first-order vector angular polarized light field is used when setting the modulation, the polarization direction coincides with the darkest direction, and the detected polarization direction is 115°. Figure 8 The white solid line represents the final detected polarization angle, which is completely consistent with the true value of 115°.

[0092] In order to test the effectiveness and robustness of the algorithm, polarized light intensity images with different polarization directions are generated in the range of 0° to 175° at intervals of 5° according to Malus's law, and Gaussian noise with a mean of 0.01 and a variance of 0.01 is added to the image to simulate the effects of shot noise, thermal noise and dark current noise in the actual imaging process. The test image is processed using three methods: global Radon transform (GR), small angle range image correlation based on a rough estimate of the region of interest (IAD+LC), and image grayscale signature feature matching (CE+GS) based on a rough estimate of multi-circular data provided by the present invention. The error of the algorithm is measured by the deviation between the estimated angle and the true angle. The estimation errors of the three methods are given by Fig. 9 The horizontal axis is the actual image polarization (darkest) direction angle, and the vertical axis is the algorithm error at this angle. Fig. 9 It shows that for images with different polarization directions, the angle extraction errors of IAD+LC and CE+GS are in the range of [0.008, 0.0228] and [0.01, 0.022], respectively, while the angle detection error of the GR algorithm is in the range of [0.7915, 1.7610]. This shows that the detection accuracy of IAD+LC and CE+GS is comparable, and both are much higher than GR.

[0093] In terms of time consumption, the time required for processing a polarization image with a pixel size of 1601×1601 is given in Table 1. The comparison shows that the speed of CE+GS is significantly improved compared with GR and IAD+LC.

[0094] Table 1. Time required for each algorithm

[0095] method GR IAD+LC CE+GS Time(s) 112.437 2.065 0.0251

[0096] The time consumption of GR, IAD+LC and CE+GS are 112.437s, 1.412s and 0.0251s respectively. In order to test the actual effect of the method provided by the present invention, a system was built to collect the modulated image of the light wave in real time. In order to test the stability and robustness of the algorithm, the polarization state of the incident light wave was kept unchanged in the experiment, and the digital camera was made to collect one image every 2 seconds. 20 actual images were collected continuously for calculation and analysis, one of which is shown in Fig.10 The three algorithms are used to measure the image, and the results are shown in Fig.11 Given. Fig.11 In the figure, the horizontal axis is the number of detections, and the vertical axis is the measured polarization direction angle. Fig.11 It can be found that the CE+GS detection result curve proposed by the present invention is the flattest. Furthermore, Table 2 shows the mean square error of the results of each algorithm, which once again verifies the robustness of CE+GS.

[0097] Table 2 Mean square error of real image measurement results

[0098] algorithm GR IAD+LC CE+GS Mean square error 0.0415 <![CDATA[5.71e -4 ]]> <![CDATA[2.63e -5 ]]>

[0099] The method provided by the present invention has the advantages of both high precision and high speed, and can fully meet the task requirements of real-time detection of polarization direction based on vector polarized light field.

[0100] Although the embodiments of the present invention are described in conjunction with the accompanying drawings, the patent owner may make various variations or modifications within the scope of the appended claims. As long as they do not exceed the protection scope described in the claims of the present invention, they should be within the protection scope of the present invention.

Claims

1. A method for rapid detection of light wave polarization direction based on image grayscale signature, characterized in that: The following steps are involved: S1, offline operation, used to generate a circumferential pixel coordinate table, a radial pixel coordinate table, a grayscale weight table and a template grayscale signature; S2, online operation, for making a rough estimate of the darkest direction according to the circumferential pixel coordinate table, quickly obtaining the grayscale signature of the polarization image according to the rough estimate of the darkest direction, the radial pixel coordinate table and the grayscale weight table, and correlating the grayscale signature of the polarization image with the grayscale signature of the template to obtain an accurate estimate of the polarization direction; The step of correlating the polarization image grayscale signature with the template grayscale signature to obtain an accurate estimate of the polarization direction in step S2 specifically includes: The image grayscale signature g is respectively compared with the template grayscale signature corresponding to each deviation angle obtained in the offline operation Perform correlation and search for the deviation angle corresponding to the template grayscale signature with the maximum value θ rc With θ sc The deviation angle range and interval set by the algorithm are respectively used to obtain an accurate estimate of the darkest direction. If the first-order vector angular polarized light field is used for modulation, the darkest direction angle is the polarization direction of the polarized light; if the first-order vector radial polarized light field is used for modulation, the polarization direction of the incident light is perpendicular to the darkest direction, that is, the polarization direction is arg(ξ) returns the parameter that satisfies the condition ξ, θ min is a rough estimate of the darkest direction angle.

2. The method for rapid detection of light wave polarization direction based on image grayscale signature according to claim 1 is characterized in that: The steps of generating the circumferential pixel coordinate table in step S1 are: For an M×N image, a polar coordinate system is established with the center pixel of the image [round(M / 2), round(N / 2)] as the origin, and the polar coordinates (r p ,θ p ); Set a set of sampling radii {R1, R2, ..., R n }, for each sampling radius R i , i = 1, 2, ..., n, if the pixel point p satisfies R i -r ξ ≤r p ≤R i +r ξ , then the pixel coordinates (u p ,v p ) is stored in radius R i The corresponding circular pixel coordinate table; Among them, round(ξ) means rounding the data ξ to the nearest integer, r ξ It is the radius error range when searching for circular pixels.

3. The method for rapid detection of light wave polarization direction based on image grayscale signature according to claim 1, characterized in that: The steps of generating the radial pixel coordinate table and the grayscale weight table in step S1 are: The image is within the radius range [R min ,R max Each pixel point p(u p ,v p ) is divided into 4 sub-pixels, whose pixel coordinates are (u p -0.25,v p -0.25)、(u p -0.25,v p +0.25)、(u p +0.25,v p -0.25) and (u p +0.25,v p +0.25); In the angle range [0°, 180°), the angle θ st For each sampling angle Draw a straight line through the origin Calculate each sub-pixel p s (u ps ,v ps ) to the straight line Vertical distance If any sub-pixel p s (u ps ,v ps ) to the straight line The vertical distance is less than the set threshold d th , then it is considered that the pixel p(u p ,v p ) is included in the above straight line, and the pixel coordinates (u p ,v p ) recorded in In the corresponding radial pixel table, the weight w of the pixel point is p Recorded in In the corresponding grayscale weight table; Among them, (x ps ,y ps ) is the sub-pixel p s (u ps ,v ps ), u0 and v0 represent the matrix row and column coordinates of the image center p0, respectively. [ξ] represents the largest integer not greater than ξ, belongs to pixel p and to the straight line The distance is less than the threshold d th The sum of all sub-pixel weights.

4. The method for rapid detection of light wave polarization direction based on image grayscale signature according to claim 1, characterized in that: The method for generating the template grayscale signature in step S1 comprises the following steps: With the pixel point with pixel coordinates of [round(M / 2),round(N / 2)] as the center, generate an ideal polarization image with the darkest direction of 0° and size of M×N as a template; The angle coverage and sampling interval of the grayscale signature feature are set to ±θ ri With θ si , with 0° as the center, for the i-th sampling angle Query the radial pixel coordinate table and grayscale weight table to obtain the radial grayscale weighted integral value in, Indicates the sampling angle The corresponding radial pixel coordinate table, I p (u j ,v j ), w p (u j ,v j ) are pixels p(u j ,v j ) and the gray value of the pixel at angle θ si The corresponding weight value in the grayscale weight table; The grayscale radial integral values ​​calculated at each sampling angle are arranged in counterclockwise order and normalized to eliminate the influence of ambient brightness to obtain the template grayscale signature.

5. The method for rapid detection of light wave polarization direction based on image grayscale signature according to claim 1, characterized in that: In the method for generating the template grayscale signature feature, if the influence of noise is taken into account: the deviation angle range is set to ±θ rc With interval θ sc , for the i-th deviation angle by The grayscale signature feature of the template is calculated as the center, and % represents the remainder operation.

6. The method for rapid detection of light wave polarization direction based on image grayscale signature according to claim 1, characterized in that: The steps of roughly estimating the darkest direction according to the circumferential pixel coordinate table in step S2 are: For the polarization image I0 to be measured, find the pixel coordinates ([u0, v0]) of the center p0 of the region in which the grayscale is wedge-shaped distributed in the image; For each sampling radius R set in the offline operation in step S1 i , i = 1, 2, ..., n, query the circular pixel coordinate table corresponding to each sampling radius, and according to each coordinate [u ri ,v ri ], extract pixel coordinates from image I0 as [u0,v0]+[u ri ,v ri ]-[round(M / 2),round(N / 2)] pixel grayscale values, and record the grayscale values ​​of the pixels in the circle that are less than the threshold value T g The pixel polar angle; Calculate the average value θ of the polar angles of the pixels extracted from each circle min The sampling angle θ is set when the radial pixel coordinate table is generated in the offline operation in step S1. st is the precision versus the mean value θ min Round off to get a rough estimate of the darkest direction angle.

7. The method for rapid detection of light wave polarization direction based on image grayscale signature according to claim 1, characterized in that: The step of obtaining the grayscale signature of the polarization image to be measured according to the rough estimated value of the darkest direction, the radial pixel coordinate table and the grayscale weight table in step S2 specifically includes: Centered on the roughly estimated darkest polarization direction angle, the grayscale signature covers an angle range of ±θ ri and the sampling interval θ si , query the radial pixel coordinate table and grayscale weight table corresponding to each sampling angle; According to the data obtained from the radial pixel coordinate table [u i ,v i ], extract the pixel coordinates of the image to be measured I0 as [u0,v0]+[u i ,v i ]-[round(M / 2),round(N / 2)] pixel grayscale, and combined with the pixel weights obtained from the grayscale weight table, calculate the grayscale signature of the polarization image to be measured; Where [u0, v0] is the pixel coordinate of the center p0 of the grayscale wedge-shaped distribution area in the polarization image I0 to be measured, M and N are the number of rows and columns of the image matrix, respectively, and round() represents the rounding operation of the data in the brackets.

Images