Polarization navigation method in night shielding environment

Abstract

The invention discloses a polarization navigation method in a night shielding environment, which specifically comprises the following three parts: building a space-time sum enhancement model, extracting feature points and fitting: in a space-time sum denoising and enhancement stage, smoothly processing background noise by adopting spatial filtering, processing bipolar impulse noise by utilizing time filtering, and extracting the feature points; enhancing the image by adopting a dung beetle-like visual nerve summation enhancement method; in the feature point extraction stage, the polarization degree and the polarization angle are calculated by using the enhanced image, an effective area is extracted, and feature points are searched; in the fitting stage, a sun meridian fitting method with asymmetrical distribution of positive and negative polarization angles based on PD adjustment is utilized, and the problems of sun meridian fitting deviation and poor orientation precision caused by asymmetry of positive and negative feature points are solved. The method can effectively improve the orientation precision in the sparse polarization state in the weak polarization shielding mode at night.

Description

Technical Field

[0001] The invention belongs to the technical field of polarized light navigation, and in particular relates to a polarized light navigation method in a nighttime shielded environment. Background Art

[0002] High-precision navigation sensors are an important guarantee for unmanned motion platforms to complete tasks autonomously, and "inertial + satellite" combined navigation is currently the most commonly used navigation method. Inertial navigation systems are prone to drift during long flight times, and satellite signals are susceptible to natural or human interference. Recent biological studies have shown that in nature, sand ants and dung beetles have the ability to use atmospheric polarization patterns for navigation. The polarization-sensitive units in the compound eye structure of sand ants can obtain atmospheric polarization patterns with high precision to achieve navigation solutions; dung beetles rely on the polarization visual neural pathway to enhance the perception system to perceive the dim moonlight at night to determine their position and achieve linear movement, which provides a new solution for autonomous navigation without relying on satellites.

[0003] Daytime polarization navigation has been widely studied in autonomous navigation, but nighttime polarization has received relatively little attention. The weak polarization light field at night poses a challenge to the collection of available navigation information. The photon emission noise is relatively larger at night than in the daytime environment, resulting in a low signal-to-noise ratio of photons at night, which is easily interfered by noise. In addition, the light intensity at night is more than 6-8 orders of magnitude lower than that during the day. Long exposure time is required to achieve the same orientation effect as during the day, but the sampling frequency is severely reduced. In addition, the bionic polarization compass also faces the problem of missing local polarization information due to obstructions such as trees and buildings during actual application. Especially under the coupling of weak polarization modes at night, the polarization orientation accuracy drops sharply. Summary of the invention

[0004] Purpose of the invention: In order to solve the problems of large noise, long exposure time and low orientation accuracy of polarization navigation images in night-time obstruction environments in the prior art, the present invention provides a polarization navigation method in night-time obstruction environments.

[0005] Technical solution: A polarization navigation method in a nighttime obstruction environment, comprising the following steps:

[0006] Step 1: Establish a spatiotemporal sum enhancement model, including image denoising and image enhancement processing;

[0007] Step 2: Calculate the polarization degree and polarization angle of the enhanced image using the Stokes vector method, extract the effective area according to the polarization degree, and find the positive feature points and negative feature points of the effective area according to the polarization angle;

[0008] Step 3: Divide the enhanced image into m equally spaced circular rings, continuously adjust the axis position, and calculate the weighted difference sum of the positive and negative feature points in the effective area of ​​the m circular rings under different axes. The axis corresponding to the minimum value of the weighted difference sum is the solar azimuth in the carrier coordinate system; obtain the solar azimuth in the navigation coordinate system through the astronomical calendar; calculate the difference between the solar azimuth in the carrier coordinate system and the navigation coordinate system to obtain the absolute heading angle.

[0009] Furthermore, in step one, the image denoising process includes: performing double filtering denoising process on the polarization image to obtain a denoised image, including using spatial filtering smoothing process to remove background noise and using time filtering to remove bipolar pulse noise.

[0010] Furthermore, in step one, the method for image enhancement processing includes: inputting a plurality of consecutive denoised images into a plurality of visual channels respectively, and accumulating the signals input from the plurality of visual channels into a single cell having a wide dendritic field, thereby enhancing the perception of light.

[0011] Furthermore, in step 2, the method for extracting the effective area according to the polarization degree includes: obtaining the axis L by fitting the feature points, taking the zenith as the center of the circle and the axis L as the baseline, rotating 30° to the left and right, dividing the image into six areas, and marking them in clockwise order, the first area, the fourth area, the third area, and the sixth area are effective areas, and the second area and the fifth area are interference areas; assuming the polarization angle of the feature point to be β, setting the polarization angle threshold α, selecting the feature points, defining the feature points whose polarization angles satisfy α≤β≤90° as positive feature points, and defining the feature points whose polarization angles satisfy -α≥β≥-90° as negative feature points.

[0012] Furthermore, in step 3, the weighted difference sum is assumed to be δ, and the calculation formula of δ is:

[0013]

[0014] Among them, P i_r_eff is the number of positive feature points in the effective area of ​​the i-th ring, N i_r_eff is the number of negative feature points in the effective area of ​​the i-th ring, P i_r is the number of positive feature points in the i-th ring, N i_r is the number of negative feature points in the ith ring, ω represents the symmetry weight of positive and negative feature points in the effective area of ​​the ith ring, P eff is the positive feature point in the effective area, N eff It is a negative feature point in the valid area.

[0015] Furthermore, in step 3, the method for determining the minimum value of the weighted difference and δ is:

[0016] The control axis L rotates by an angle of θ each time, automatically adjusting the accuracy and computational complexity, so that it can quickly find the axis position corresponding to the minimum value of δ.

[0017] θ(k+1)=K p *δ(k)+K d *(δ(k)-δ(k-1)) (3)

[0018] Among them, K p is the proportionality coefficient, K d is the differential coefficient, k is the kth rotation axis L.

[0019] Compared with the prior art, the polarization navigation method in a night-time obstructed environment provided by the present invention has the following beneficial effects: it solves the problem of poor orientation accuracy in a night-time obstructed environment, proposes a double denoising process for the high noise of polarization images collected at night, adopts spatial filtering to smooth the background noise, and uses time filtering to process the bipolar pulse noise; adopts a dung beetle visual nerve summation enhancement method to enhance the image; no longer has high requirements for exposure time; selects effective areas to extract feature points in a regional manner to reduce interference from obstruction factors, and uses a solar meridian fitting method with asymmetric distribution of positive and negative polarization angles based on PD adjustment to solve the problem of solar meridian offset and poor orientation accuracy caused by asymmetric positive and negative feature points, improves the solution speed, and effectively increases the environmental adaptability of night-time polarization navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a polarization navigation method in a nighttime occlusion environment;

[0021] Figure 2 This is the orientation result of the building shading experiment in the weak polarization mode at night;

[0022] Figure 3 This is the orientation result of the tree shading experiment in the nighttime weak polarization mode;

[0023] Figure 4 It is the box error diagram of the building occlusion and tree occlusion experiments in the weak polarization mode at night. DETAILED DESCRIPTION

[0024] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0025] A polarization navigation method in a nighttime occlusion environment, such as Figure 1 As shown, the following steps are included:

[0026] Step 1: Establish a spatiotemporal summation enhancement model, specifically including: performing double filtering denoising on the polarization image to obtain a denoised image, including using spatial filtering smoothing to remove background noise and using temporal filtering to remove bipolar pulse noise; and enhancing the denoised image using a dung beetle visual nerve summation enhancement method to obtain an enhanced image.

[0027] Assume that the atmospheric polarization pattern signal under ideal conditions is x(t), the electrical noise and channel noise are classified as Gaussian white noise n1, and the weather noise makes the image brightness distribution uneven and produces bipolar impulse noise n2(t), so the polarization image can be expressed as:

[0028] x i1 (t) = x(t) + n1 + n2(t) (1)

[0029] The background noise n1 is smoothed by spatial filtering. The spatial filter K(n) is expressed as:

[0030]

[0031] Where σ represents the standard deviation of the spatial filter and n is the exponent;

[0032] The coefficients of the spatial filter are obtained through the second-order derivative,

[0033]

[0034] Normalize the filter:

[0035]

[0036] In order to achieve less ripples and narrower filtering, the stacking coefficient c is added for regulation to filter out Gaussian white noise, as shown in formula (5):

[0037]

[0038] The method of removing bipolar pulse noise by time filtering is:

[0039] First, decompose the impulse noise n2(t) to obtain the M-order noise, and weight the M-order noise to obtain the first-order modal component:

[0040]

[0041] Calculate the residual signal r1(t):

[0042] r1(t)=n2(t)-E1(t) (7)

[0043] The residual signal r1(t) is decomposed M times using formula (6) to obtain the second-order modal components. The decomposition of each modal component is repeated multiple times to express the original signal as follows:

[0044]

[0045] The method for enhancing the denoised image includes: inputting multiple denoised images into multiple visual channels, and accumulating them into a single cell with a wide dendritic field, thereby enhancing the perception of light. (Inputting multiple images into multiple channels, accumulating the outputs of multiple channels, and achieving light enhancement)

[0046] Step 2: Use the Stokes vector method to calculate the polarization degree and polarization angle of the enhanced image, segment the blocked area according to the polarization degree to extract the effective area, and find the feature points near the absolute value of the polarization angle of 90° according to the polarization angle.

[0047] The specific method for extracting the effective area includes: obtaining the axis L by fitting the feature points, taking the zenith point as the center of the circle and the axis L as the baseline, rotating 30° left and right, dividing the image into six areas, and marking them in clockwise order, the first area, the fourth area, the third area, and the sixth area are effective areas, and the second area and the fifth area are interference areas; setting the polarization angle of the feature point as β, setting the polarization angle threshold α, selecting the feature point, defining the feature point whose polarization angle satisfies α≤β≤90° as a positive feature point, and defining the feature point whose polarization angle satisfies -α≥β≥-90° as a negative feature point.

[0048] Step 3: Divide the enhanced image into m equally spaced circular rings, continuously adjust the axis position, and calculate the weighted difference and δ of the positive and negative feature points in the effective area of ​​the m circular rings under different axes. The axis corresponding to the minimum value of the weighted difference is the solar azimuth in the carrier coordinate system; obtain the solar azimuth in the navigation coordinate system through the astronomical almanac; calculate the difference between the solar azimuths in the carrier coordinate system and the navigation coordinate system to obtain the absolute heading angle.

[0049] Assume the sum of weighted differences is δ, and the calculation formula of δ is:

[0050]

[0051] Among them, D i_r_eff is the difference between the positive and negative feature points in the effective area of ​​the i-th ring, ω ip_r_eff is the probability of a positive feature point in the i-th ring, P i_r is the number of positive feature points in the i-th ring, ω in_r_eff is the probability of negative feature points in the i-th ring, N i_r is the number of negative feature points in the i-th ring, ω is the weight factor, and ω is inversely proportional to the difference between the number of positive and negative feature points in the effective area;

[0052] D i_r_eff =P i_r_eff -N i_r_eff (10)

[0053] Among them, P i_r_eff is the number of positive feature points in the effective area of ​​the i-th ring, N i_r_eff is the number of negative feature points in the valid area of ​​the i-th ring;

[0054]

[0055] The formula (9), (11), (12) can be simplified to:

[0056]

[0057] P eff is the positive feature point in the effective area, N eff is the negative feature point of the effective area, and ω represents the symmetry weight of the positive and negative feature points of the effective area in the i-th ring.

[0058] The method for determining the minimum value of the weighted difference and δ is:

[0059] The control axis L rotates by an angle of θ each time, automatically adjusting the accuracy and computational complexity, so that it can quickly find the axis position corresponding to the minimum value of δ.

[0060] θ(k+1)=K p *δ(k)+K d *(δ(k)-δ(k-1)) (15)

[0061] Among them, K p is the proportionality coefficient, K d is the differential coefficient, k refers to the kth rotation axis L.

[0062] In order to verify the accuracy of the polarization navigation method of this embodiment under nighttime occlusion environment, the following experiments were conducted to demonstrate: Figure 2 , Figure 3 The experimental results of the building occlusion and the tree occlusion in the polarization mode at night are respectively compared with the orientation results of the least square method and the benchmark value. Figure 4 The above are box error diagrams of two experiments. It can be seen that the heading angle obtained by this method has higher accuracy and smaller error.

Claims

1. A polarization navigation method in a nighttime obstruction environment, characterized in that: The following steps are involved: Step 1: Establish a spatiotemporal sum enhancement model, including image denoising and image enhancement processing; Step 2: Calculate the polarization degree and polarization angle of the enhanced image using the Stokes vector method, extract the effective area according to the polarization degree, and find the positive feature points and negative feature points of the effective area according to the polarization angle; Step 3: Divide the enhanced image into m equally spaced circular rings, continuously adjust the axis position, and calculate the weighted difference sum of the positive and negative feature points in the effective area of ​​the m circular rings under different axes. The axis corresponding to the minimum value of the weighted difference sum is the solar azimuth in the carrier coordinate system. The solar azimuth in the navigation coordinate system is obtained through the astronomical almanac; the difference between the solar azimuth in the carrier coordinate system and the navigation coordinate system is calculated to obtain the absolute heading angle.

2. The polarization navigation method in a nighttime obstruction environment according to claim 1, characterized in that: In step 1, the image denoising process includes: performing double filtering denoising process on the polarization image to obtain a denoised image, including using spatial filtering smoothing process to remove background noise and using time filtering to remove bipolar pulse noise.

3. The polarization navigation method in a nighttime obstruction environment according to claim 1 or 2, characterized in that: In step 1, the method for image enhancement processing includes: inputting a plurality of consecutive denoised images into a plurality of visual channels respectively, and accumulating the signals input from the plurality of visual channels into a single cell with a wide dendritic field, thereby enhancing the perception of light.

4. The polarization navigation method in a nighttime obstruction environment according to claim 1 or 2, characterized in that: In step 2, the method for extracting the effective area according to the polarization degree includes: obtaining the axis L by fitting the feature points, taking the zenith as the center of the circle and the axis L as the baseline, rotating 30° to the left and right, dividing the image into six areas, and marking them in clockwise order, the first area, the fourth area, the third area, and the sixth area are effective areas, and the second area and the fifth area are interference areas; assuming the polarization angle of the feature point to be β, setting the polarization angle threshold α, selecting the feature point, defining the feature point whose polarization angle satisfies α≤β≤90° as a positive feature point, and defining the feature point whose polarization angle satisfies -α≥β≥-90° as a negative feature point.

5. The polarization navigation method in a nighttime obstruction environment according to claim 1 or 2, characterized in that: In step 3, let the sum of weighted differences be δ, and the calculation formula of δ is: Among them, P i_r_eff is the number of positive feature points in the effective area of ​​the i-th ring, N i_r_eff is the number of negative feature points in the effective area of ​​the i-th ring, P i_r is the number of positive feature points in the i-th ring, N i_r is the number of negative feature points in the ith ring, ω represents the symmetry weight of positive and negative feature points in the effective area of ​​the ith ring, P eff is the positive feature point in the effective area, N eff It is a negative feature point in the valid area.

6. The polarization navigation method in a nighttime obstruction environment according to claim 1 or 2, characterized in that: In step 3, the method for determining the minimum value of the weighted difference and δ is: The control axis L rotates by an angle of θ each time, automatically adjusting the accuracy and computational complexity, so that it can quickly find the axis position corresponding to the minimum value of δ. θ(k+1)=K p *δ(k)+K d *(δ(k)-δ(k-1)) (3) Among them, K p is the proportionality coefficient, K d is the differential coefficient, k is the kth rotation axis L.

Images