Polarization orientation method based on sky area prior and transform domain morphological template matching
Through a polarization orientation method based on sky region prior and transform domain morphological template matching, the navigation technology is solved to local polarization anomalies and solar fuzzy problems, and high-precision heading angle measurement and robust navigation under reflected light interference are achieved.
Patent Information
- Application Number
- CN202410080283.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-18
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-01-18
AI Technical Summary
The existing navigation technology is extremely sensitive to local polarization anomalies, resulting in large calculation errors. Direct use of polarization angle navigation will cause solar fuzzy problems, increasing navigation costs.
Using a polarization orientation method based on the matching of the a priori and transform domain morphological templates of the sky region, the polarization dark channel image and gradient image were calculated by the polarization camera, clustering and feature extraction were performed, and the sun's position was determined in combination with the spatial domain transformation method, and Rayleigh was ignored. Scattering model difference, calculate heading angle.
Under the condition of reflected light interference, the heading angle measurement error can be achieved, and the sun's blur can be judged with high robustness, and it does not rely on additional sensors, which improves navigation accuracy and robustness.
Smart Images

Figure CN118010010B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of navigation technology, and in particular relates to a polarization orientation method based on sky area priori and transform domain morphological template matching. Background Art
[0002] Studies have shown that the angle of polarization (AOP) predicted by the single Rayleigh scattering model is robust and contains a large amount of navigation information, which can complete accurate heading angle measurements. However, current orientation methods are often sensitive to local polarization anomalies, which will increase calculation errors, especially when reflected light from objects such as trees and buildings destroys the AOP model. Although the prior art proposes methods such as polarization gradient, it should be noted that this method cannot identify obstructions with flat characteristics. Another key issue is that directly using the angle of polarization (AOP) for navigation will cause solar ambiguity problems, because the range of polarization angles is only half of the range of heading angles. Therefore, this method requires additional information to detect solar ambiguity, which undoubtedly increases the navigation cost. Summary of the invention
[0003] Purpose of the invention: In order to solve the problem that the orientation method in the prior art is interfered by reflected light and has poor accuracy, and needs to increase navigation cost to solve the problem of solar blur, the present invention provides a polarization orientation method based on sky area prior and transform domain morphological template matching.
[0004] Technical solution: A polarization orientation method based on sky area prior and transform domain morphological template matching, comprising the following steps:
[0005] (1) Using the 0°, 45°, 90° and 135° images taken by the polarization camera, the polarization dark channel image, polarization degree, polarization degree gradient and polarization angle gradient are calculated; the light intensity binary image and the occlusion binary image are calculated based on the polarization dark channel image; the polarization degree binary image is calculated based on the polarization degree; and the gradient binary image is calculated based on the polarization degree gradient and the polarization angle gradient;
[0006] (2) Setting the number of clusters according to the light intensity binary image and the occlusion binary image;
[0007] (3) performing polarization dark channel clustering according to a set number of clusters to obtain a polarization dark channel clustering graph, and obtaining a sky area image through the polarization dark channel clustering graph;
[0008] (4) calculating the polarization angle image and performing feature extraction on the polarization angle image to obtain a polarization angle feature binary image;
[0009] (5) constructing a two-dimensional plane with (0°, 360°] and [-90°, 90°] as the horizontal direction and the vertical direction respectively, and projecting the polarization angle feature binary image onto the two-dimensional plane using a spatial domain transformation method to obtain a binary feature map of the sun position;
[0010] (6) The sun position and the reverse sun position are obtained by matching the sun position binary feature map, and the real sun position is determined according to the constraint that the sun is above the horizontal plane;
[0011] (7) Ignore the difference between the Rayleigh scattering model and the real atmosphere model, calculate the sun position deviation, and if the deviation is less than the threshold T m The test result is considered to be true and reliable; if the deviation is greater than the threshold, the test result is considered to be unreliable;
[0012] (8) Calculate the heading angle by combining the polarization angle and the sky area image.
[0013] Furthermore, in step (1), the polarization dark channel is expressed as:
[0014]
[0015] where C = [r, g, b, r0, g0, b0, r 45 , g 45 , b 45 , r 90 , g 90 , b 90 , r 135 , g 135 , b 135 ]
[0016] Among them, r, g, and b represent the polarization images of the red, green, and blue channels, respectively. * Representative *-angle red channel polarization image.
[0017] Furthermore, in step (1), the light intensity I H , polarization degree I DOP 、Occlusion I S And the binary image of gradient ΔI is expressed as:
[0018] I H =1(I D =I max ),
[0019] I DOP =1(1%<P D <90%)
[0020] I S =1(I D =0)
[0021] ΔI=1((ΔP D +ΔP A )>T)
[0022] Among them, I D Represents each element in the polarized dark channel image matrix, I max Indicates the maximum light intensity that the sensor can accept, P D Denotes the degree of polarization, ΔP D represents the polarization gradient, ΔP A represents the polarization angle gradient, T is the gradient threshold, and 1 represents the indicator function.
[0023] Furthermore, in step (2), the method for setting the number of clusters is:
[0024] (a) Set the number of clusters k o The initial value of is 1;
[0025] (b) If there is occlusion, update the number of clusters to:
[0026]
[0027] (c) Let the intensity threshold be wNM. If the light intensity exceeds wNM, the number of clusters obtained in step (b) is further updated as:
[0028]
[0029] Furthermore, in step (3), the sky area image Ω is obtained through the polarization dark channel clustering map SC The acquisition method is:
[0030] Ω SC =μ sc ∩ΔI∩I H ∩I S
[0031] Among them, μ sc Represents the sky class, μ sc =I K (μ s =0,μ h =0,μ m =0)
[0032] Among them, I K represents the polarization dark channel clustering diagram, μ s For occlusion class, μ h is the intensity class, μ m is the fuzzy class,
[0033]
[0034] Among them, μ represents the number of clusters ko The updated set of all categories, μ k represents the kth class, k∈(1, 2, ...K0), N k is the total number of individuals in the kth class.
[0035] Furthermore, in step (4), the polarization angle feature binary image AOP b0 The method to obtain is:
[0036] AOP b0 (find(ρ=cosh s cos(θ-A s )))=1,
[0037] Among them, find(*) represents the coordinate set that satisfies *.
[0038]
[0039] Among them, h s is the solar altitude angle, h p is the altitude angle of the observation point, A p is the azimuth of the observation point.
[0040] Furthermore, in step (6), the sun position and the reverse sun position are obtained by matching the sun position binary feature map:
[0041]
[0042] Add constraints:
[0043] ξ t =ξ,stsign(ξ)>0
[0044] Among them, ξ represents the sun's position, + is the sun's position on the horizontal plane, ξ- is the sun's position below the horizontal plane, is the solar altitude angle on the horizontal plane, The solar azimuth on the horizontal plane, The sun's altitude angle below the horizon, The solar azimuth below the horizontal plane, ξ t Represents the true sun position;
[0045] According to the sun position, the inverse sun position and the constraints, the real sun position is obtained, denoted as ξ t (h s , A s ).
[0046] Further, in step (7), the sun position deviation is calculated as:
[0047]
[0048] Among them, δ represents the deviation of the sun’s position, δ b0 Represents the polarization angle feature binary image AOP b0 The deviation, represents cos(A p -A s ), Representative of coth p , A s is the solar azimuth.
[0049] Furthermore, in step (8), the heading angle for:
[0050]
[0051] Among them, θ nb is the pitch angle, φ nb is the roll angle, measured by an additional sensor,
[0052]
[0053] in,
[0054]
[0055] Among them, x, V, z are defined as
[0056]
[0057] Where NM is the total number of pixels, e ij To measure the polarization angle, {i, j|Ω SC (i, j)==1}.
[0058] Beneficial effect: Compared with the existing technology, the polarization orientation method based on sky area prior and transform domain morphological template matching provided by the present invention can realize heading angle measurement under the condition of reflected light interference from objects such as trees and buildings, with small error, and can complete highly robust solar ambiguity judgment without relying on additional sensors, without requiring additional costs, and effectively improves the accuracy and robustness of the heading angle measurement method based on the atmospheric polarization pattern.
[0059] Compared with the existing orientation method based on atmospheric scattered light, this method visualizes the navigation information and improves the robustness of the orientation. This method provides a polarization angle transformation domain model, introduces template matching to resolve solar blur on binary feature images, and gives an accurate sky area based on the polarization dark channel and regional prior before orientation, thus realizing the measurement of heading angle when part of the sky scattered light is destroyed by reflected light. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 It is a framework flow chart of the polarization orientation method based on sky area prior and transform domain morphological template matching;
[0061] Figure 2 The schematic diagram of the sun blur judgment result using the polarization orientation method based on sky area prior and transform domain morphological template matching;
[0062] Figure 3 Schematic diagram of the orientation error measurement results using the polarization orientation method based on sky area prior and transform domain morphological template matching. DETAILED DESCRIPTION
[0063] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.
[0064] A polarization orientation method based on sky area prior and transform domain morphological template matching, such as Figure 1 As shown, the following steps are included:
[0065] (1) The polarization dark channel image, polarization degree, polarization degree gradient and polarization angle gradient are calculated using the 0°, 45°, 90° and 135° images taken by the polarization camera. The light intensity binary image and occlusion binary image are calculated based on the polarization dark channel image. The polarization degree binary image is calculated based on the polarization degree. The gradient binary image is calculated based on the polarization degree gradient and polarization angle gradient. The details are as follows:
[0066] Light intensity I H , polarization degree I DOP 、Occlusion I S And the binary image of gradient ΔI is expressed as:
[0067] I H =1(I D =I max ),
[0068] I DOP =1(1%<P D <90%)
[0069] I S =1(I D =0)
[0070] ΔI=1((ΔP D +ΔP A )>T)
[0071] Among them, I D Represents each element in the polarized dark channel image matrix, I max Indicates the maximum light intensity that the sensor can accept, P D Denotes the degree of polarization, ΔP Drepresents the polarization gradient, ΔP A represents the polarization angle gradient, T is the gradient threshold, and 1 represents the indicator function.
[0072] The polarized dark channel image is represented as:
[0073]
[0074] where C = [r, g, b, r0, g0, b0, r 45 , g 45 , b 45 , r 90 , g 90 , b 90 , r 135 , g 135 , b 135 ]
[0075] Among them, r, g, and b represent the polarization images of the red, green, and blue channels, respectively. * Representative *-angle red channel polarization image.
[0076] (2) Setting the number of clusters according to the light intensity binary image and the occlusion binary image;
[0077] The method to set the number of clusters is:
[0078] (a) Set the number of clusters k o The initial value of is 1;
[0079] (b) If there is occlusion, update the number of clusters to:
[0080]
[0081] (c) Let the intensity threshold be wNM. If the light intensity exceeds wNM, the number of clusters obtained in step (b) is further updated as:
[0082]
[0083] (3) Perform polarization dark channel clustering according to the set cluster number to obtain a polarization dark channel cluster map, and obtain a sky area image through the polarization dark channel cluster map. The acquisition method is as follows:
[0084] Ω SC =μ sc ∩ΔI∩I H ∩I S
[0085] Among them, μ sc Represents the sky class, μ sc =I K (μ s =0,μ h=0,μ m =0)
[0086] Among them, I K represents the polarization dark channel clustering diagram, μ s For occlusion class, μ h is the intensity class, μ m Fuzzy class.
[0087]
[0088] Among them, μ represents the number of clusters k o The updated set of all categories, μ k represents the kth class, k∈{1, 2, …k0}, N k is the total number of individuals in the kth class.
[0089] (4) Calculate the polarization angle image, extract features from the polarization angle image, and obtain a polarization angle feature binary image, which is expressed as:
[0090] AOP b0 (find(ρ=cosh s cos(θ-A s )))=1,
[0091] Among them, find(*) represents the coordinate set that satisfies *.
[0092]
[0093] Among them, h s is the solar altitude angle, h p is the altitude angle of the observation point, A p is the azimuth of the observation point.
[0094] (5) constructing a two-dimensional plane with (0°, 360°] and [-90°, 90°] as the horizontal direction and the vertical direction respectively, and projecting the polarization angle feature binary image onto the two-dimensional plane using a spatial domain transformation method to obtain a binary feature map of the sun position;
[0095] (6) The sun position and the reverse sun position are obtained by matching the sun position binary feature map:
[0096]
[0097] Since the sun is above the horizon, set the constraint condition.
[0098] ξ t =ξ,stsign(ξ)>0
[0099] Among them, ξ represents the sun's position, +is the sun's position on the horizontal plane, ξ- is the sun's position below the horizontal plane, is the solar altitude angle on the horizontal plane, The solar azimuth on the horizontal plane, The sun's altitude angle below the horizon, The solar azimuth below the horizon, ξ t Represents the true sun position;
[0100] According to the sun position, the inverse sun position and the constraints, the real sun position is obtained, denoted as ξ t (h s , A s ).
[0101] (7) Ignore the difference between the Rayleigh scattering model and the real atmosphere model, calculate the sun position deviation, and if the deviation is less than the threshold T m The test result is considered to be true and reliable; if the deviation is greater than the threshold, the test result is considered unreliable and can be re-tested and calculated later.
[0102] The sun position deviation is calculated as:
[0103]
[0104] Among them, δ represents the deviation of the sun’s position, δ b0 Represents the polarization angle feature binary image AOP b0 The deviation, represents cos(A p -A s ), Representative of coth p , A s is the solar altitude angle. b0 Proportional to pass The standard deviation of represents the deviation in the calculation of the sun's position, which can be simplified as
[0105] (8) Calculate the heading angle by combining the polarization angle and the sky area image. for:
[0106]
[0107] Among them, θ nb is the pitch angle, φ nb is the roll angle. The pitch angle and roll angle are measured by additional sensors such as an inertial navigation system.
[0108]
[0109] in,
[0110]
[0111] Among them, x, y, z are defined as
[0112]
[0113] Where NM is the total number of pixels, e ij To measure the polarization angle, {i, j|Ω SC (i, j)==1}.
[0114] The effect of this method is verified by vehicle-mounted tests. Figure 2 Schematic diagram of the sun blur judgment result of this method. Figure 3 The following is a comparison of vehicle orientation errors of different polarization orientation methods. The orientation methods used for comparison include: traditional orientation using polarized light directly, orientation using polarization gradient (Fan), and orientation using weather weighted sparse coding (WWSC). Figure 3 It can be seen that the method of this embodiment has the highest accuracy under the condition of reflected light interference and exhibits good robustness under different conditions.
[0115] This method gives the criteria for region division through the prior knowledge of the sky area, and completes the recognition of the sky area by clustering the darkest pixel images in the 15-channel polarization images. At the same time, the proposed transform domain morphological template matching (MTMTD) method processes the polarization angle image in the transform domain through an imaging method, solving the problem of solar blur under the single Rayleigh scattering model without relying on additional sensors. This method can complete orientation under ground / near-ground conditions when the sky conditions are obstructed, effectively improving the accuracy and robustness of heading measurement based on atmospheric polarized light.
Claims
1. A polarization orientation method based on sky area prior and transform domain morphological template matching, It is characterized in that The following steps are involved: (1) Using the 0°, 45°, 90° and 135° images taken by the polarization camera, the polarization dark channel image, polarization degree, polarization degree gradient and polarization angle gradient are calculated; the light intensity binary image and the occlusion binary image are calculated based on the polarization dark channel image; the polarization degree binary image is calculated based on the polarization degree; and the gradient binary image is calculated based on the polarization degree gradient and the polarization angle gradient; Among them, the light intensity I H , polarization degree I DOP 、Occlusion I S And the binary image of gradient ΔI is expressed as: I H =1(I D =I max ), I DOP =1(1%<P D <90%) I S =1(I D =0) ΔI=1((ΔP D +ΔP A )>T) Among them, I D Represents each element in the polarized dark channel image matrix, I max Indicates the maximum light intensity that the sensor can accept, P D Denotes the degree of polarization, ΔP D represents the polarization gradient, ΔP A represents the polarization angle gradient, T is the gradient threshold, and 1 represents the indicator function; (2) Setting the number of clusters according to the light intensity binary image and the occlusion binary image; (3) Perform polarization dark channel clustering according to the set clustering number to obtain a polarization dark channel clustering graph, and obtain the sky area image Ω through the polarization dark channel clustering graph SC , the specific acquisition method is: Oh SC =μ sc ∩ΔI∩I H ∩I S Among them, μ sc Represents the sky class, μ sc =I K (μ s =0,μ h =0,μ m =0) Among them, I K represents the polarization dark channel clustering diagram, μ s For occlusion class, μ h is the intensity class, μ m is the fuzzy class, Among them, μ represents the number of clusters k o The updated set of all categories, μ k represents the kth class, k∈{1, 2, …K 0 },N k is the total number of individuals in the kth class; (4) calculating the polarization angle image and performing feature extraction on the polarization angle image to obtain a polarization angle feature binary image; (5) constructing a two-dimensional plane with (0°, 360°] and [-90°, 90°] as the horizontal direction and the vertical direction respectively, and projecting the polarization angle feature binary image onto the two-dimensional plane using a spatial domain transformation method to obtain a binary feature map of the sun position; (6) The sun position and the reverse sun position are obtained by matching the sun position binary feature map: Add constraints: x t =ξ,stsign(ξ)>0 Among them, ξ represents the sun's position, + is the sun's position on the horizontal plane, ξ - is the sun's position below the horizon, is the solar altitude angle on the horizontal plane, The solar azimuth on the horizontal plane, The sun's altitude angle below the horizon, The solar azimuth below the horizontal plane, ξ t Represents the true sun position; According to the sun position and anti-sun position and the constraint that the sun is above the horizontal plane, the real sun position is obtained, denoted as ξ t (h s , A s ); (7) Ignore the difference between the Rayleigh scattering model and the real atmosphere model, calculate the sun position deviation, and if the deviation is less than the threshold T m The test result is considered to be true and reliable; if the deviation is greater than the threshold, the test result is considered to be unreliable; (8) Calculate the heading angle by combining the polarization angle and the sky area image. for: Among them, θ nb is the pitch angle, φ nb is the roll angle, measured by an additional sensor, in, Where x, y, and z are defined as: Where NM is the total number of pixels, e ij To measure the polarization angle, {i, j|Ω SC (i, j)==1}.
2. The polarization orientation method based on sky area prior and transform domain morphological template matching according to claim 1, It is characterized in that In step (1), the polarization dark channel is expressed as: where C = [r, g, b, r 0 , g 0 , b 0 , r 45 , g 45 , b 45 , r 90 , g 90 , b 90 , r 135 , g 135 , b 135 Among them, r, g, and b represent the polarization images of the red, green, and blue channels, respectively. * Representative *-angle red channel polarization image.
3. The polarization orientation method based on sky area prior and transform domain morphological template matching according to claim 1 or 2, It is characterized in that In step (2), the method for setting the number of clusters is: (a) Set the number of clusters k o The initial value of is 1; (b) If there is occlusion, update the number of clusters to: (c) Let the intensity threshold be wNM. If the light intensity exceeds wNM, the number of clusters obtained in step (b) is further updated as:
4. The polarization orientation method based on sky area prior and transform domain morphological template matching according to claim 1 or 2, It is characterized in that In step (4), the polarization angle feature binary image AOP b0 The method to obtain is: AOP b0 (find(ρ=cosh s cos(θ-A s )))=1, Among them, find(*) represents the coordinate set that satisfies *. Among them, h s is the solar altitude angle, h p is the altitude angle of the observation point, A p is the azimuth of the observation point, A s is the solar azimuth.
5. The polarization orientation method based on sky area prior and transform domain morphological template matching according to claim 1 or 2, It is characterized in that In step (7), the sun position deviation is calculated as: Among them, δ represents the deviation of the sun’s position, δ b0 Represents the polarization angle feature binary image AOP b0 The deviation, represents cos(A p -A s ), Representative of coth p , A s is the solar azimuth.
