Method for absolute depth measurement based on parallax angle of polarization microlens light field image

Abstract

The present application relates to the technical field of light field imaging measurement, and particularly relates to an absolute depth measurement method based on a parallax angle of a polarized microlens light field image. The method comprises the following steps: 1) image pixel level classification based on multi-view sub-aperture depth maps, which provides a benchmark for optimization of a parallax map; 2) constructing a matching cost equation based on pixel classification results, so as to realize global optimization of a relative depth map from coarse to fine; 3) establishing a linear mapping model from the relative depth map to the absolute depth map based on the hardware structure of a microlens light field camera, so as to realize depth calibration; 4) combining the depth calibration with internal parameters of the light field camera, so as to calculate absolute depth physical parameters of an observed target. Thus, physical measurement and characterization of three-dimensional absolute size of a real observed object are realized, and absolute depth information of a target to be measured in a light field image can be obtained, in particular, absolute depth information of a weak texture and a blocked target can be obtained.

Description

Technical Field

[0001] This invention relates to the field of light field imaging measurement technology, and in particular to an absolute depth measurement method based on the parallax angle of a polarized microlens light field image. Background Technology

[0002] With the continuous development of society and the economy, the demand for 3D remote sensing information is increasing, making the rapid, accurate, and economical acquisition of 3D information about ground features increasingly important. Traditional remote sensing imaging equipment uses digital cameras, which consist of optical lenses and electronic photosensitive elements. Their imaging process converts the 3D information of an object into a 2D planar image. Reconstructing the surface information of a 3D object from the image is the reverse process, requiring multiple images taken from different angles by a digital camera, and then using software calculations and processing to reconstruct the object's 3D information. Because a significant amount of information is lost during the projection from the 3D world to the 2D image, the uncertainty of the reverse process is inherent. Furthermore, 3D reconstruction is computationally complex, and its accuracy is affected by various factors, severely limiting the acquisition of 3D remote sensing information.

[0003] With the advent of light field imaging technology, microlens light field cameras can now acquire information about the direction of light within the camera. This characteristic allows for the acquisition of 3D information of a target from a single light field image, providing a new approach to obtaining 3D scene information. Compared to traditional camera imaging methods, light field imaging offers more efficient and realistic 3D reconstruction. For a single light field camera image, the key to stereo matching parallax calculation based on sub-aperture images is finding the corresponding projection points in the left and right images. This process requires consideration of accuracy and processing speed, making pixel matching crucial in the entire stereo matching process. If there are occluded areas or weak textures in the light field image, traditional depth extraction methods based on stereo matching may experience mismatches. Therefore, it is essential to employ effective methods to reduce mismatches.

[0004] To obtain the absolute depth information of the target, the extracted relative depth information must be calibrated. This invention uses a first-generation microlens light field camera as its hardware foundation and calculates the absolute depth based on the parallax angle of the light field image acquired by the camera. After constructing the depth map of the target object by classifying the image pixels, the optical structure of the camera is analyzed, simplifying the depth calibration of the microlens array light field camera to linear calibration. The depth information is converted into real three-dimensional information by using the camera intrinsic parameter calibration and depth calibration, and then applied to the three-dimensional measurement of real objects, providing a new method for measuring the real three-dimensional information of targets in the field of light fields. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an absolute depth measurement method based on the parallax angle of a light field image using a polarizing microlens, which can acquire the absolute depth information of the target to be measured in the light field image, especially the absolute depth information of weak texture and occluded targets.

[0006] To address the aforementioned problems, the present invention proposes the following technical solution: an absolute depth measurement method based on the parallax angle of a polarizing microlens light field image, comprising:

[0007] Step 1: Image pixel-level classification based on multi-view sub-aperture depth maps provides a benchmark for the optimization of disparity maps;

[0008] Step 2: Construct a matching cost equation based on pixel classification results to achieve global optimization of the relative depth map from coarse to fine.

[0009] Step 3: Based on the hardware structure of the microlens light field camera, establish a linear mapping model from the relative depth map to the absolute depth map to achieve depth calibration.

[0010] Step four: Combine the depth calibration with the intrinsic parameters of the light field camera to calculate the absolute depth physical parameters of the observed target.

[0011] This enables the physical measurement and representation of the three-dimensional absolute size of real observed objects.

[0012] First, all pixels in the image are classified. For pixels with occlusion and weak texture in the polarization microlens light field image, the above conventional method will result in an ill-conditioned matching cost equation. The depth corresponding to the minimum matching cost is often not the correct depth. This phenomenon is called mismatch. It is difficult to eliminate mismatch, but a more reasonable depth value can be obtained by using the correct depth of non-occluded and non-weak textured pixels. For pixels without occlusion, the depth information of the pixels in the center view should be consistent with that of the corresponding points in other view. However, for pixels with occlusion, the pixels in the center view image cannot find the corresponding point information in the images of other view using depth value matching retrieval. When the depth information of the center view image is inconsistent with that of other pixels, such pixels are defined as occluded pixels as shown in equation (1):

[0013] depth c (p c )≠depth i (p c +(n c -n i )) (1)

[0014] Where depth c It is a depth map of the central sub-aperture image. iIt is the depth map of other sub-aperture images, p c n represents any point in the central sub-aperture image. c -n i This represents the distance between the central sub-aperture and other sub-aperture images.

[0015] In the light field sub-aperture image array, different sub-aperture images represent different observation directions, and therefore the occlusion parts are different. This patent first calculates the unoccluded pixels in the reference image based on the depth map of each sub-aperture image, and then merges them together. In areas with weak texture, because the distinguishability of pixels is poor, the constructed matching cost equation does not have an obvious minimum value. Therefore, if the minimum and the second minimum values ​​are small, it indicates that the pixel texture is weak. First, the depth of the corresponding points of the pixels is compared to distinguish the occluded pixels and the unoccluded pixels. Then, the difference between the second minimum and the minimum matching cost of the unoccluded pixels is calculated. The pixels are divided into stable pixels and weak texture pixels. The pixels of this type are defined as weak texture pixels as shown in equation (2):

[0016]

[0017] in This represents the minimum value in the matching cost equation at a certain point. It is the second smallest value in the matching cost function, τ textless Represents the threshold.

[0018] See the correlation of pixel classification in light field images Figure 1 As shown. First, the depths of corresponding points are compared to distinguish occluded and non-occluded pixels. Then, calculations are performed on the non-occluded pixels to classify them into weak-texture pixels and strong-texture pixels. For weak-texture pixels, since the pixel is not easily distinguishable from its surrounding pixels, there is no obvious minimum value when constructing the matching cost. Therefore, if the minimum and second-minimum values ​​of the matching cost of a pixel are small, it indicates that the pixel is a weak-texture pixel.

[0019] A matching cost equation based on pixel classification results is constructed. After classifying all pixels into three different pixel types using a pixel classification method, different cost functions E are constructed using multi-view correlation to construct the matching cost equation, as shown in equation (3):

[0020]

[0021] in Let I(p,c) represent the initial matching cost equation, V represent the set of all viewpoints of the light field image, v represent the image angle, p represent the image space coordinates (x,y), median represents the median function, and min represents the minimum function. The image from the center viewpoint is denoted as I(p,c), while the image from any viewpoint is denoted as I(p,c).

[0022] Secondly, based on the aforementioned function cost equation, a method for obtaining the light field depth is used, progressing from coarse to fine and from local to global for weak textures and occluded pixels. The main steps of this matching cost optimization method are as follows:

[0023] 1) A pixel classification method is used to divide all pixels into three different pixel types. Sub-aperture arrays obtained by decoding the light field image are used to construct different cost functions based on multi-view correlation. 2) A matching cost function is constructed based on the initial stereo matching results. An adaptive window is used to optimize the cost information, thereby establishing a relative depth map. 3) After minimizing the cost value, weighted median filtering is used to remove noise from the depth map, thereby obtaining an optimized relative depth map. 4) The relative depth map is segmented into a patch structure. RANSAC (Random Sample Consensus) plane fitting is used for each patch. The depth information of the entire patch is fitted based on the relatively accurate depth information of stable pixels. The fitted depth information is then used to correct the matching cost equation. Multiple iterations are performed to obtain the globally optimal relative depth map.

[0024] After an image is segmented into a patch structure, the depth of pixels within each patch is basically the same. Therefore, unstable pixels can be approximated using the depth information of stable pixels. Specifically, RANSAC plane fitting is first performed on each patch, using the coordinates and depth (u, v, depth) of stable pixels as the fitting data.

[0025] Different scaling factors are used to correct the initial matching cost equation according to the different pixel categories, as shown in formula (4). For occluded pixels, since no matching pixel can be found, the matching cost equation must be constructed using the results obtained from fitting stable pixels in the patch. For weak texture pixels, the initial matching cost and the fitted matching cost are considered together. For stable pixels, the initial matching cost equation is considered to be accurate, and the influence of patch fitting depth is less considered. The fitting result of all data is calculated, and then the random selection of subsets is repeated to calculate and update the new fitting result and error. Compared with the previous fitting result, the fitting result with smaller error is retained, and the iteration continues until the convergence condition is reached. Let E represent the initial matching cost equation for the i-th iteration, and E D This represents the matching cost equation obtained by the WTA (Winner Takes All) method.

[0026]

[0027] For details on the optimization method, please refer to Figure 2 .

[0028] Furthermore, for the specific three-dimensional reconstruction method of light field images, please refer to [link / reference]. Figure 3 Using polarized microlens light field images to achieve true geometric measurement of the target object, after acquiring the depth map, camera calibration and depth calibration are still required. A checkerboard calibration method is used in the intrinsic parameter calibration process, and the calibrated parameters are the virtual camera intrinsic parameters. The intrinsic parameters of the decomposed image method are the intrinsic parameters of the virtual camera corresponding to the central image; therefore, there is no need to consider the complex microlens image calibration process.

[0029] Thirdly, for the depth calibration process, linear calibration is employed to avoid calibration errors caused by overfitting or underfitting. Simultaneously, geometric dimension measurements on a checkerboard pattern combined with camera intrinsic parameter calculations are used to reduce systematic errors. The following is the derivation process of the linear relationship between depth distance information and parallax information. The optical structure of the microlens light field camera is shown below. Figure 4 :

[0030] Where P represents the object point, P′ represents the image point, b represents the distance from the microlens array to the imaging plane, B represents the distance between the primary lens and the microlens array, d represents the aperture of a single microlens, D represents the aperture of the primary lens, z represents the distance from the image point to the microlens array, and Z represents the distance from the object point to the primary lens. Relative depth information is obtained through stereo matching, and the following formula (5) is derived based on the formula for calculating the polar parallax angle:

[0031]

[0032] Where depth0 represents the parallax, Δθ represents the parallax angle, and ρ represents the distance from the image point to the imaging plane.

[0033] According to the Gaussian imaging formula, we obtain equation (6):

[0034]

[0035] Where f2 is the focal length of the primary lens. Further, we obtain equation (7):

[0036]

[0037] The design of the light field camera satisfies the F# number matching principle, that is, the F# numbers of the primary lens and the microlens are equal, therefore equation (8) holds:

[0038]

[0039] Combining equations (7) and (8), we obtain equation (9):

[0040]

[0041] Because the diameter of the microlens is much smaller than the diameter of the main lens, therefore:

[0042]

[0043] Equation (10) can be written as:

[0044]

[0045] Solving for c1 and c2 completes the depth calibration. Specifically, a light field camera takes a picture at point Z from a fixed position. Depth0 at Z is calculated using depth extraction. After obtaining a sufficient number of data sets, the least squares method is used to calculate the linear equation of the scatter plot of depth0 versus 1 / Z, thus determining c1 and c2. Therefore, when performing depth calibration on multi-angle image sets using a light field camera, a linear calibration method can be employed.

[0046] Finally, the light field image of the checkerboard pattern was captured from multiple angles. The intrinsic parameter matrix K was determined according to the intrinsic parameter calibration method of the light field camera. At the same time, the absolute depth was determined by combining the relative depth map and the calibration curve. The absolute three-dimensional information of the captured target was further calculated. Each pixel in the image has parameters (u, v, depth). The calculation method of the three-dimensional spatial coordinates is shown in Equation (12):

[0047]

[0048] Since the camera is used for shooting, the camera coordinate system is chosen as the world coordinate system. The relationship between pixel coordinates and 3D coordinates is shown in equation (13).

[0049]

[0050] After camera calibration, the camera's internal parameters were determined to be as follows: so:

[0051]

[0052] The beneficial effects of this invention are:

[0053] This invention relates to an absolute depth measurement method for the parallax angle of a polarized microlens light field image, which can acquire the absolute depth information of the target in the light field image, especially the absolute depth information of weak texture and occluded targets. First, the image depth map is acquired using pixel classification, and an optimized cost cube is constructed. A high-precision relative depth map is constructed using the polar coordinate parallax angle calculation formula. Furthermore, starting from the structure of the light field camera, an absolute depth calibration model is derived and established.

[0054] This invention is the first to perform linear calibration of relative depth information to achieve absolute depth measurement of target objects in light field images. Simultaneously, starting from images captured by a first-generation light field camera, a depth map of the target object is constructed through 3D reconstruction using the acquired light field information. Then, camera intrinsic parameter calibration and depth calibration are used to convert the depth map information into true 3D information, thereby completing the absolute depth measurement of the target object. This provides a new method for practical depth measurement in the field of light field imaging. Attached Figure Description

[0055] Figure 1 This is a graph showing the pixel classification and correlation of the light field image in the absolute depth measurement method based on the parallax angle of the light field image of the present invention.

[0056] Figure 2 This is a schematic diagram illustrating the optimization of the light field image matching cost in the absolute depth measurement method based on the parallax angle of the light field image of the present invention.

[0057] Figure 3 This is a schematic diagram of the three-dimensional reconstruction method of the light field image based on the absolute depth measurement method of the parallax angle of the light field image of the present invention.

[0058] Figure 4 This is a schematic diagram of the optical structure of a microlens light field camera for measuring the absolute depth of light field images based on the parallax angle of polarized microlens light field images, according to the present invention.

[0059] Figure 5 This is a schematic diagram illustrating the extraction and optimization of the relative depth map of the target key from the light field image of the absolute depth measurement method based on the parallax angle of the light field image of the polarized microlens light field image according to the present invention.

[0060] Figure 6 This is a schematic diagram showing the positional relationship between the camera and the calibration plate in the light field image of the absolute depth measurement method based on the parallax angle of the light field image of the polarization microlens light field image according to the present invention.

[0061] Figure 7 This is a schematic diagram of the linear fitting result of the depth calibration of the light field image of the absolute depth measurement method based on the parallax angle of the light field image of the present invention.

[0062] Figure 8 This is a schematic diagram of a multi-angle grid image for camera intrinsic parameter calibration of the light field image of the absolute depth measurement method based on the parallax angle of the light field image of the present invention.

[0063] Figure 9 This is a schematic diagram of the measurement of a scene key in a light field image based on the absolute depth measurement method of the parallax angle of a polarized microlens light field image according to the present invention. Detailed Implementation

[0064] The invention will now be further described with reference to the accompanying drawings.

[0065] The absolute depth measurement method based on the parallax angle of a polarizing microlens light field image provided by this invention includes the following steps:

[0066] Step 1: Image pixel-level classification based on multi-view sub-aperture depth maps provides a benchmark for the optimization of disparity maps;

[0067] Step 2: Construct a matching cost equation based on pixel classification results to achieve global optimization of the relative depth map from coarse to fine.

[0068] Step 3: Based on the hardware structure of the microlens light field camera, establish a linear mapping model from the relative depth map to the absolute depth map to achieve depth calibration.

[0069] Step four: Combine the depth calibration with the intrinsic parameters of the light field camera to calculate the absolute depth physical parameters of the observed target.

[0070] The optimization benchmark for establishing the disparity map in step one involves obtaining the image depth map using image pixel-level classification. Based on the multi-view sub-aperture depth map, pixels are divided into occluded pixels, weakly textured pixels, and stable pixels, ensuring that the extracted depth maps are consistent across local objects. The image is then segmented into a patch structure, and the occluded pixels in the image are obtained using formula (1).

[0071] depth c (p c )≠depth i (p c +(n c -n i )) (1)

[0072] Where depth c It is a depth map of the central sub-aperture image. n It is the depth map of other sub-aperture images, p c n represents any point in the central sub-aperture image. c -n i The distance between the central sub-aperture and other sub-aperture images is represented by , while the weak texture pixels are obtained by formula (2): This represents the minimum value of the matching cost function at a certain point. It is the second smallest value of the matching cost function, τ textless Represents the threshold.

[0073]

[0074] The matching cost equation in step two uses a pixel classification method to divide all pixels into three different pixel types. Different cost functions are constructed using multi-view correlation. The matching cost equation is determined by formula (3):

[0075]

[0076] Where V represents the set of all viewpoints of the light field image, v represents the image angle, p represents the image space coordinates (x, y), median represents the median function, and min represents the minimum function. The image from the central viewpoint is denoted as I(p, c), while the image from any viewpoint is denoted as I(p, c). This represents the initial matching cost equation constructed.

[0077] In step two, the global optimization of the relative depth map utilizes multi-view correlation to construct different cost functions, and uses an adaptive window to optimize cost information to establish the relative depth map. After noise removal using weighted median filtering, the relative depth map is divided into patch structures. For each patch, RANSAC plane fitting is used to fit the depth information of the entire patch based on the accurate depth information of stable pixels, and the fitted depth information is used to correct the matching cost equation. This process is repeated multiple times to obtain the globally optimal relative depth map, where the iteration formula is shown in (4). Let E represent the initial matching cost equation for the i-th iteration, and E D This represents the matching cost equation obtained by the WTA (Winner Takes All) method.

[0078]

[0079] The absolute depth calibration model of the light field image in step three obtains relative depth information through stereo matching. Based on the polar coordinate parallax angle calculation formula, the following equation (5) is obtained:

[0080]

[0081] Where depth0 represents the parallax, Δθ represents the parallax angle, and ρ represents the distance from the image point to the imaging plane.

[0082] The method obtained, in measuring relative depth information, derives and establishes a linear relationship between depth distance information and parallax information based on the hardware structure of the microlens camera, and completes an accurate depth calibration curve using a small number of points. Equation (6) holds true:

[0083]

[0084] Where f2 is the focal length of the primary lens, and depth0 represents the parallax. Combining the fact that the F# of the primary lens and the microlens are equal, the linear model of the absolute depth Z and the relative depth is obtained as shown in formula (7):

[0085]

[0086] In step three, depth calibration involves using a microlens light field camera to vertically capture images of the checkerboard pattern. Relative depth information is extracted from the checkerboard images at different depths, while simultaneously recording the actual distance. After repeatedly acquiring multiple sets of data, a scatter plot of the relative depth information and the reciprocal of the actual distance is created. Simultaneously, the least squares method is used for linear fitting to determine the specific parameters c1 and c2.

[0087] The physical parameters of the absolute depth of the observed target in step four. A checkerboard light field image is captured from multiple angles. The intrinsic parameter matrix K is determined according to the intrinsic parameter calibration method of the light field camera. Simultaneously, the absolute depth is determined by combining the relative depth map and the calibration curve. Further calculations are then performed to obtain the absolute three-dimensional information of the captured target. Each pixel in the image has parameters (u, v, depth). The formulas for calculating the three-dimensional spatial coordinates are formulas (8) and (9):

[0088]

[0089]

[0090] Working principle:

[0091] First, pixel classification is used to acquire the relative depth map of the image, and an optimized cost cube is constructed to complete the construction of a high-precision depth map. Figure 5 The following are presented: (a) the light field center image of the experimental scene of the target under test; (b) the depth map optimized by cost matching function using pixel classification and adaptive window; (c) the noise removal of the depth map by weighted median filtering after minimizing the cost value, thus obtaining the optimized relative depth map; (d) the relative depth map is segmented into a patch structure; RANSAC plane fitting is used for each patch, and the depth information of the entire patch is fitted based on the relatively accurate depth information of stable pixels. The matching cost equation is then corrected using the fitted depth information, and the global optimal relative depth map is obtained through multiple iterations.

[0092] Furthermore, based on the camera structure of the microlens light field camera, a linear mapping relationship for absolute depth calibration is established. The specific method for depth calibration is to use the microlens light field camera to vertically capture the checkerboard image, such as... Figure 6 As shown, the distance Z from the camera to the chessboard grid is in the range [6, 26], in centimeters. A total of 21 sets of light field data were captured. Depth information was extracted from each image. To improve the accuracy of image depth extraction, the average depth of pixels at the edges of the entire chessboard grid was extracted. According to Formula 10, a scatter plot of depth versus the reciprocal of distance was obtained. Linear fitting was performed using the least squares method. The linear fitting results for depth calibration are shown in […]. Figure 7 .

[0093] After completing the depth calibration, nine light field images of a checkerboard pattern were taken from multiple angles using Zhang Zhengyou's calibration method. Since the central viewpoint was used as the reference image during depth calculation, the sub-aperture images of the central viewpoint were calibrated using intrinsic parameters.

[0094] Feature points are extracted from each image, and the homography matrix is ​​calculated. Then, the intrinsic parameter matrix is ​​solved using the homography matrices of multiple images. The calculated intrinsic parameter matrix of the light field camera is as follows:

[0095]

[0096] After the relative depth map is obtained, each pixel in the image has parameters (u, v, depth0). After depth calibration using the method described in the introduction, the absolute depth Z is calculated according to Formula 15. W .

[0097]

[0098] After camera calibration, the camera's intrinsic parameters were determined to be... Get (X) W Y W ):

[0099]

[0100] Table 1. Experimental measurement results for scenario one.

[0101]

[0102]

[0103] Figure 9 As shown, in scene 1, the coordinates of the upper left point A of the key are [75, 149], and the relative depth value (parallax value) is 42. Based on the depth calibration results in Figure 7, Z is calculated... WThen, X is calculated based on the camera's intrinsic parameters. W and Y W .

[0104]

[0105]

[0106] Table 1 shows the experimental results for Scenario 1. Similarly, the 3D coordinates of point A in Scenario 1 are (-3.2216, -1.4170, 21.7391), and the coordinates of point B in Scenario 1 are (2.3824, 0.3552, 21.7391). The Euclidean distance between them is 5.8775cm. This was measured multiple times using vernier calipers in Scene 1. The distance between them is 5.686 cm, so the error is 0.1915 cm. The results are shown in Table 1. The method of the present invention can accurately calculate the three-dimensional information of the target in the image.

[0107] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the claims.

Claims

1. An absolute depth measurement method based on the parallax angle of a polarizing microlens light field image, characterized in that: Includes the following steps: Step 1: Image pixel-level classification based on multi-view sub-aperture depth maps provides a benchmark for the optimization of disparity maps; Step 2: Construct a matching cost equation based on pixel classification results to achieve global optimization of the relative depth map from coarse to fine. Step 3: Based on the hardware structure of the microlens light field camera, establish a linear mapping model from the relative depth map to the absolute depth map to achieve depth calibration. Step four: Combine the depth calibration with the intrinsic parameters of the light field camera to calculate the absolute depth physical parameters of the observed target; The matching cost equation in step two uses a pixel classification method to divide all pixels into three different pixel types. Different cost functions are constructed using multi-view correlation. The matching cost equation is determined by formula (3): （3） Where V represents the set of all viewpoints of the light field image, v represents the image angle, p represents the image spatial coordinates (x, y), median represents the median function, and min represents the minimum function. Let I(p,c) represent the initial matching cost equation, I(p,v) represent the gray value of pixel p in the central sub-aperture c, I(p,v) represent the gray value of pixel p in the sub-aperture v, and N(v) is the number of sub-apertures contained in the neighborhood sub-aperture set, which is an integer greater than 0.

2. The absolute depth measurement method based on the parallax angle of a polarizing microlens light field image according to claim 1, characterized in that: The optimization benchmark for establishing the disparity map in step one involves obtaining the image depth map using image pixel-level classification. Based on the multi-view sub-aperture depth map, pixels are divided into occluded pixels, weakly textured pixels, and stable pixels, ensuring that the extracted depth maps are consistent across local objects. The image is then segmented into a patch structure, and the occluded pixels in the image are obtained using formula (1). （1） Where depth c (p c ) represents pixel p in the central sub-aperture c. c depth value i (p c +(n c -n i )) represents the pixel p in the i-th sub-aperture that is closest to the central sub-aperture. c The corresponding depth value of the same point; where n c Let be the index value of the central sub-aperture, and ni be the index value of the i-th non-central sub-aperture, where i is the sequence number of the non-central sub-aperture, and i = 1, 2, ..., k, where k is the total number of non-central sub-apertures. c -n i ) represents the index difference between the central sub-aperture and the i-th sub-aperture, used to determine the offset position of the corresponding point, while the weak texture pixel is obtained by formula (2): This represents the minimum value of the matching cost function at a certain point. It is the second smallest value of the matching cost function. Represents the threshold. （2）。 3. The absolute depth measurement method based on the parallax angle of a polarizing microlens light field image according to claim 2, characterized in that: In step two, the global optimization of the relative depth map involves constructing different types of cost functions using multi-view correlation, optimizing the cost functions using an adaptive window to establish the relative depth map, removing noise using weighted median filtering, and then segmenting the relative depth map into patch structures. For each patch, RANSAC plane fitting is used to fit the depth information of the entire patch based on the accurate depth information of stable pixels, and the fitted depth information is used to correct the matching cost equation. This process is repeated multiple times to obtain the globally optimal relative depth map, where the iteration formula is shown in (4). Let represent the initial matching cost equation for the i-th iteration, and The matching cost equation obtained by the WTA (Winner Takes All) method is as follows: （4）。 4. The absolute depth measurement method based on the parallax angle of a polarized microlens light field image according to claim 3, characterized in that: In step three, the absolute depth calibration obtains relative depth information through stereo matching. A linear mapping model is established based on the hardware structure of the microlens light field camera, and the following equation (5) is obtained according to the polar coordinate disparity angle calculation formula: （5） in Indicates parallax. Represents the parallax angle. f1 represents the distance from the image point to the imaging plane, z represents the distance from the image point to the microlens array, and f1 is the focal length of the microlens.

5. The absolute depth measurement method based on the parallax angle of a polarized microlens light field image according to claim 4, characterized in that: In the measurement of relative depth information, starting from the hardware structure of the microlens camera, a linear relationship between depth distance information and parallax information is derived and established. An accurate depth calibration curve is completed using a small number of points, and formula (6) holds true: （6） Where f2 is the focal length of the primary lens and depth0 represents the parallax, and combining the fact that the F# of the primary lens and the microlens are equal, the linear model of the absolute depth Z and the relative depth is obtained as shown in formula (7): （7） Where Z is the absolute depth, B is the baseline distance, and c1 and c2 are the weighting coefficients.

6. The absolute depth measurement method based on the parallax angle of a polarized microlens light field image according to claim 5, characterized in that: In step three, the depth calibration involves using a microlens light field camera to vertically capture the chessboard image and extracting relative depth information from the chessboard images at different depths. Simultaneously, the actual distance is recorded. After repeatedly acquiring multiple sets of data, a scatter plot of relative depth information and the reciprocal of the actual distance is established. At the same time, the least squares method is used for linear fitting to determine the specific parameters c1 and c2.

7. The absolute depth measurement method based on the parallax angle of a polarizing microlens light field image according to claim 6, characterized in that: The physical parameters of the absolute depth of the observed target in step four are obtained by capturing light field images of a checkerboard pattern from multiple angles. The intrinsic parameter matrix K is determined according to the intrinsic parameter calibration method of the light field camera. At the same time, the absolute depth is determined by combining the relative depth map and the calibration curve. The absolute three-dimensional information of the captured target is then calculated, and parameters are obtained for each pixel in the image. The formulas for calculating the three-dimensional coordinates in space are formulas (8) and (9): （8） （9）。

Images