A method for 3D reconstruction of objects based on shadow image processing

Abstract

The present invention discloses a method for target three-dimensional reconstruction based on shadow image processing, comprising the following steps: obtaining a target shadow image; preprocessing the shadow image; calculating the target motion direction; building a target model; transforming coordinate information; and removing redundant points from the target model to obtain a target three-dimensional reconstruction result. The present invention uses a sequence of shadow images of a moving target at different angles to restore the target's three-dimensional structure, does not rely on high-precision registration of target feature points, has low algorithm complexity, and can achieve real-time processing.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a target three-dimensional reconstruction method based on shadow image processing. Background Art

[0002] Due to the frequency band advantage of terahertz SAR images, shadows in SAR images can be used for the detection, tracking and positioning of moving targets, and the shadow shape can also reflect the shape information of the target projected onto the ground by the radar wave. Therefore, in SAR images in the terahertz frequency band, it is possible to classify or identify moving targets by shadow size, shape and other features. The formation of target shadows in SAR images is similar to that in optical images, which is caused by visible light or electromagnetic waves being blocked by the target. When the target's moving direction changes at a certain angle relative to the radar's line of sight, the moving target can form shadows of different shapes on the ground. Therefore, using shadow information to reconstruct moving targets in three dimensions can integrate shadow information at multiple angles or postures, obtain more information, and provide new ideas for moving target recognition.

[0003] The existing method of using image sequences to perform three-dimensional reconstruction of targets is as follows:

[0004] Deng Baosong disclosed a method for 3D reconstruction of target models in "Research on 3D Reconstruction Technology of Large Baseline Image Sequences Based on Point and Line Features". This method uses point and line features to recover the surface topological mesh and texture information of the target model from a large baseline optical image sequence.

[0005] Liu Xiaolong disclosed a target 3D reconstruction method in "Research on Theory and Methods of Euclidean 3D Reconstruction Based on Image Sequences". This method requires manual calibration, image registration and accurate camera posture. It uses the Euclidean 3D reconstruction method to reconstruct the optical target and has high robustness.

[0006] Cui Dejing disclosed a three-dimensional reconstruction method for moving objects in "Research on Three-Dimensional Reconstruction Method of Moving Targets Based on Image Sequences". This method uses time-frequency characteristics to improve the accuracy of image registration, and uses the improved triangulation method to build a three-dimensional grid model for discrete points to reconstruct the three-dimensional target.

[0007] In short, most of the existing 3D reconstruction methods based on image sequences are based on optical image sequences and require high-precision registration. However, the target shadows in SAR images only have contour features, which do not meet the requirements of 3D reconstruction of optical images. Therefore, it is necessary to study the 3D reconstruction method of moving targets based on shadow information in SAR image sequences. Summary of the invention

[0008] In order to solve the above problems, the present invention proposes a target three-dimensional reconstruction method based on shadow image processing.

[0009] Obtaining a shadow image of a target, wherein the target is a moving target, and the shadow image is a SAR image;

[0010] Preprocessing the shadow image, the preprocessing including image interpolation, image binarization and image edge smoothing, to obtain a binary image after edge smoothing;

[0011] Calculation of target movement direction: Calculate the target movement direction using the azimuth speed and distance speed of the moving target;

[0012] Constructing a target model, wherein the target model is a three-dimensional lattice model, and the three-dimensional size of the target model is larger than its actual size;

[0013] Coordinate information transformation: Based on the radar attitude angle information and the target movement direction, the target model is projected onto the ground to obtain two-dimensional coordinate information using the coordinate transformation method;

[0014] Redundant points of the target model are removed to obtain a three-dimensional reconstruction result of the target. The redundant point removal includes projection redundant point removal, prior redundant point removal and model internal point removal. The projection redundant point removal is to adjust the coordinate scale of the two-dimensional coordinate information obtained by coordinate information transformation and match it with the binary image after edge smoothing to delete the information of some points in the two-dimensional coordinate information.

[0015] Furthermore, the coordinate information conversion specifically includes the following steps:

[0016] Calculate the coordinate rotation matrix M rot ,

[0017]

[0018] Among them, θ is the target motion direction angle;

[0019] The initial target model coordinate set P is rotated to obtain the rotated coordinate set P rot ;

[0020] P rot =P×M rot

[0021] Assume that the slant range between the radar and the target is R, and the coordinates are represented by (x′, y′, z′)(x1, y1, z1). According to the radar downward viewing angle α, the radar coordinates are calculated as:

[0022]

[0023] The coordinates of the target model midpoint are expressed as (x″ i ,y″ i ,z″ i ) indicates that the projection coordinate of the point on the ground is (x″′i ,y″′ i ,z″′ i ), three points (x′, y′, z′), (x″ i ,y″ i ,z″ i ) and (x″′ i ,y″′ i ,z″′ i ) is on a straight line; for P rot For each point in the graph, calculate the coordinates of its projection point on the ground as (x″′ i ,y″′ i ,z″′ i ), the coordinate set of the target model point projection on the ground is obtained as P proj ;

[0024] in,

[0025] (x″ i ,y″ i ,z″ i ) is the rotated coordinate set P rot The coordinates of the i-th point in ;

[0026] Select multiple θ values ​​and repeat the above steps to obtain a set of projection coordinates of the target on the ground at different angles θ, that is, obtain the shadow information of the target on the ground at different angles θ.

[0027] Furthermore, the following operations are performed after the coordinate information transformation and before the redundant points are removed:

[0028] The coordinate set P of the target model point projected on the ground proj Perform scaling based on the resolution ρ, and convert P proj The unit is adjusted to pixels, and the calculation formula is as follows:

[0029] P′ proj =P proj / ρ

[0030] Furthermore, the projection redundant point elimination specifically includes the following steps:

[0031] Generate the first target point deletion mark variable D1, with all elements initialized to 0, and its elements correspond one-to-one to the points in the coordinate set P of the initial target model;

[0032] P′ proj The centroid and edge smoothing of all coordinates in the binary image I bw_s The centroids of the middle shadows are adjusted to be the same, and the binary image I is obtained after the centroid and edge are smoothed. bw_s The same point coordinate set P′proj2 ; Traverse the binary image I after edge smoothing bw_s The mid-coordinate position is P′ proj2 All pixels of the binary image I after edge smoothing bw_s If the pixel value of a certain pixel point in is 0, the corresponding element in the first target point deletion mark variable D1 is assigned a value of 1;

[0033] For P′ at multiple angles θ proj , respectively perform centroid adjustment and D1 assignment, and finally obtain the final first target point deletion mark variable D1 that meets the conditions of each angle; correspond the elements of the final first target point deletion mark variable D1 with a value of 1 to the midpoint of the initial target model coordinate set P, delete all the coordinates of the midpoint of the initial target model coordinate set P, and obtain the coordinate set P′ of the target model after the projection redundant points are eliminated;

[0034] The correspondence is performed based on the positional relationship.

[0035] Furthermore, P′ proj The specific steps for adjusting the coordinate centroid are: let P′ proj The coordinates of the center of mass are (x pc ,y pc ,z pc ), binary image I after edge smoothing bw_s The coordinates of the center of mass of the shadow are (x bwc ,y bwc ,z bwc ), P′ proj The coordinates of each point in (x bwc -x pc ,y bwc -y pc ,z bwc -z pc ) added.

[0036] Furthermore, the a priori redundant point elimination specifically includes the following steps:

[0037] Generate a second target point deletion mark variable D2, whose elements correspond one-to-one to the points in the coordinate set P of the initial target model, and the values ​​of K elements are initialized to 0;

[0038] Assuming that the target is symmetrical along the xOz plane, calculate the coordinates of the points symmetrical along the xOz plane in the coordinate set P of the initial target model, and find the target points corresponding to the coordinates of the two symmetrical points. Delete the value in the second target point deletion flag variable D2. If the values ​​are 0 and 1 respectively, assign the value 0 to 1;

[0039] Display the midpoint coordinates of the coordinate set P′ of the target model after removing redundant projection points on the width and height xOz plane, find the top triangular pyramid of the target, calculate the height coordinates of the bottom of the pyramid, find all the points whose z coordinates exceed the height coordinates of the bottom of the pyramid in the coordinate set P′ of the target model after removing redundant projection points, correspond the found points to the elements in the second target point deletion mark variable D2, and assign the corresponding elements in the second target point deletion mark variable D2 to 1;

[0040] Calculate the coordinate range of the width of the target, find all points whose y coordinates exceed the coordinate range of the width of the target in the coordinate set P′ of the target model after the projection redundant points are removed, correspond the found points to the elements in the second target point deletion mark variable D2, and assign the corresponding elements in the second target point deletion mark variable D2 to 1;

[0041] Delete all the coordinates whose corresponding second target point deletion mark variable D2 value is 1 in the coordinate set P′ of the target model after the projection redundant points are eliminated, and obtain the coordinate set P″ of the target model after the prior redundant points are eliminated;

[0042] Among them, the coordinate system where the x-axis, y-axis and z-axis are located is a three-dimensional rectangular coordinate system with the center of mass of the target as the origin and the symmetry plane of the target as the xOz plane. The x-axis, y-axis and z-axis correspond to the length, width and height of the target respectively; the correspondence is carried out according to the positional relationship.

[0043] Furthermore, the model internal point elimination specifically includes the following steps:

[0044] Generate a third target point deletion flag variable D3, initialize all elements to 1, and have a one-to-one correspondence with the points in the coordinate set P of the initial target model;

[0045] Construct a three-dimensional matrix T whose size is The values ​​of its elements are all initialized to 1, and the points in the coordinate set P″ of the target model after eliminating the prior redundant points are matched with the elements in the three-dimensional matrix T, and the corresponding elements in the three-dimensional matrix T are assigned 0, and the matching is performed according to the position relationship;

[0046] Traverse and search each element in the three-dimensional matrix T to find out all elements whose values ​​in the 3×3×3 neighborhood are all 0, correspond the found elements to the elements in the third target point deletion mark variable D3, and assign the corresponding elements in the third target point deletion mark variable D3 to 1; the correspondence is performed according to the position relationship;

[0047] The element with the element value of 1 in the third target point deletion mark variable D3 corresponds to the midpoint of the coordinate set P″ of the target model after the prior redundant points are eliminated, and the coordinates of all the corresponding points in the coordinate set P″ of the target model after the prior redundant points are eliminated are deleted to obtain the coordinate set P″′ of the target model after the internal points of the model are eliminated. The three-dimensional lattice structure formed by the coordinate set P″′ of the target model after the internal points of the model are eliminated is the three-dimensional reconstruction result of the target.

[0048] Furthermore, the target motion direction calculation specifically includes the following steps:

[0049] Assume that the direction of movement is replaced by angle information, with the radar line of sight as 0° and rotating clockwise, the target movement direction angle range is [0°, 360°);

[0050] The velocity information is smoothed using the neighborhood mean method, and the target's moving direction angle θ is calculated as:

[0051]

[0052] Among them, V r and V a They are the range velocity and azimuth velocity of the target after smoothing.

[0053] Furthermore, the shadow image preprocessing specifically includes:

[0054] Assume that the shadow image is represented by I. According to the radar viewing angle, the shadow image I is re-interpolated so that the distance information of the image is the distance information of the actual scene, and the image I is obtained. ins ;

[0055] Set the binarization threshold T bw , for image I ins After binarization, the erosion and expansion operations are performed to reduce the roughness of the shadow edge. Finally, the largest connected domain in the image is retained as the shadow area to obtain the binary image I bw ;

[0056] Extract the shadow edge in the binary image, smooth the shadow edge through smoothing operation, and then fill the inside of the edge to obtain the binary image I after edge smoothing bw_s .

[0057] Compared with the prior art, the present invention has the following significant advantages:

[0058] (1) The present invention utilizes the shadow image sequence of the moving target at different angles to restore the three-dimensional structure of the target.

[0059] (2) The present invention does not rely on the imaging of moving targets or three-dimensional imaging methods. The three-dimensional structure of the target can be restored only through the shadow information of the moving target at multiple angles, providing a new idea for the recognition of moving targets in SAR images.

[0060] (3) Compared with other target three-dimensional reconstruction methods, the present invention does not rely on high-precision registration of target feature points, has low algorithm complexity, and can achieve real-time processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a SAR image of a moving target shadow.

[0062] Figure 2 is a shadow binary image.

[0063] Figure 3 is a shadow binary edge image.

[0064] Figure 4 It is a binary edge image after edge smoothing.

[0065] Figure 5 is the moving target speed curve.

[0066] Figure 6 It is the moving target speed curve after smoothing.

[0067] Figure 7 is the target motion direction angle.

[0068] Figure 8 It is the length and width top view after removing redundant projection points.

[0069] Fig. 9 It is the height and width front view after removing redundant projection points.

[0070] Fig.10 This is the long and high side view after removing redundant projection points.

[0071] Fig.11 Delete redundant points of prior information in the height and width front view of the target model.

[0072] Fig.12 Delete redundant points of prior information from the length and width top view of the target model.

[0073] Fig.13 It is the length and width top view after removing the prior redundant points.

[0074] Fig.14 It is the height and width front view after removing the prior redundant points.

[0075] Fig.15 This is the long and tall side view after removing the a priori redundant points.

[0076] Fig.16 It is the three-dimensional dot matrix model of the vehicle after removing the internal points. DETAILED DESCRIPTION

[0077] The following is a detailed description of a specific implementation of a method for three-dimensional reconstruction of a target based on shadow image processing according to the present invention in conjunction with the accompanying drawings.

[0078] Embodiment 1

[0079] The present invention discusses the target shadow tracking results and the known target azimuth and range speeds. The target's direction of motion relative to the radar line of sight can be calculated from the target azimuth and range speeds, and then the shadow image at the corresponding moment is used to obtain the projection of the target model on the ground, and finally the redundant points in the target model are eliminated. Through the projection of the target model at multiple angles and the elimination of points, the target three-dimensional reconstruction result is obtained, which specifically includes the following steps: suppose the shadow image I, the number of image range and azimuth points are M, N respectively, in the present invention, M is the number of rows, N is the number of columns, the viewing angle under the radar is α, and the image azimuth and range resolutions are both ρ; suppose the number of shadow image I range and azimuth points are M=100, N=100 respectively, the viewing angle under the radar is α=30°, and the image azimuth and range resolutions are both ρ=0.15m.

[0080] 1. Shadow information preprocessing

[0081] Shadow information preprocessing includes three steps: shadow image interpolation, image binarization, and image edge smoothing.

[0082] Since the radar has a certain viewing angle, the range information in the shadow image reflects the distance of the radar line of sight. The shadow image I is re-interpolated so that the image's range information is the distance information of the actual scene.

[0083] Interpolate the distance of the shadow image I to get I ins , so that the number of rows and columns is Where round() means rounding to the nearest integer. Substituting the specific vertical number into [116,100], Figure 1 shown.

[0084] Set the binarization threshold T bw =0.5, for I ins Binarization is performed, where the shadow part has a value of 1 and the background part has a value of 0. Due to the influence of speckle noise, there are many small connected domains in the binary image. Morphological operations such as corrosion and expansion are performed on them to reduce the roughness of the shadow edge. Each connected domain in the image is extracted, the largest connected domain is retained, and the rest are deleted. The deletion method is to set all pixel values ​​of the connected domain to 0, so that the only connected domain in the image is the shadow area, and the binary image I is obtained. bw ,like Figure 2shown.

[0085] Since the edge of the shadow in the binary image is still relatively rough, it is necessary to smooth the shadow edge. First, extract the shadow edge in the binary image, and then smooth the edge through a smoothing operation to reduce the roughness. Figure 3 As shown. Since the shadow edge is a closed interval, the inside of the edge is filled to obtain the binary image I after edge smoothing bw_s ,like Figure 4 shown.

[0086] 2. Calculation of target movement direction

[0087] The target's range and azimuth velocities are known, and the target's moving direction can be calculated. The SAR image used in the present invention has a radar line of sight that is parallel to the paper and upward. Assuming that the moving direction is represented by angle information, with the parallel to the paper upward being 0° and rotating clockwise, the target moving direction angle range is [0°, 360°).

[0088] Since there is a certain error in the target speed estimation, the speed information between adjacent frames has a large deviation, such as Figure 5 Therefore, the neighborhood mean method is used to smooth the speed information first, and then the moving direction angle is calculated. The smoothed speed information is as follows: Figure 6 Assume that the target range velocity and azimuth velocity after smoothing are V r and V a , then the target's moving direction angle θ is:

[0089]

[0090] According to formula (1), the target's moving speed direction angle θ is calculated as follows: Figure 7 shown.

[0091] 3. Target model initialization

[0092] The moving target model is set as a three-dimensional dot matrix structure. The three-dimensional size should be larger than the actual size of the target. The length, width and height of the target model are set to L = 20m, W = 8m, and H = 4m respectively. According to the actual resolution of the SAR image and the actual requirements of three-dimensional reconstruction, the distance between points is set to r = 0.15m, and the number of points in the length direction is The number of points in the width direction is The number of points in the height direction is The target model is initialized to have a length, width, and height of (After substituting the specific values, it becomes [133,53,27]). Suppose the coordinate set of the points of the initial target model is P, P is a K×3 matrix, K represents the total number of points, 3 represents the dimension of the coordinate of each point, and the three dimensions represent the length, width and height coordinate values ​​respectively.

[0093] 4. Coordinate information transformation

[0094] During the radar imaging process, the target and the radar are both in motion, and the positional relationship between the two can be determined by the angle between the radar line of sight and the target motion direction. The SAR images used in the present invention are all irradiated from bottom to top by the radar parallel to the paper surface, so the angle between the radar line of sight and the target motion direction is θ. The coordinate information transformation is based on the positional relationship between the radar and the target, simulating the shadow of the target on the ground under the corresponding positional relationship conditions, which is defined as the target projection shadow. The coordinates of the midpoint of the target model are rotated and calculated by the target motion direction angle θ to satisfy the positional relationship between the radar and the target, and then the target projection shadow is obtained by coordinate calculation.

[0095] (1) Coordinate rotation

[0096] The coordinate rotation matrix is ​​calculated according to the target motion direction angle θ:

[0097]

[0098] The coordinates of the midpoint of the target model are rotated to obtain P rot :

[0099] P rot =P×M rot (3)

[0100] (2) Coordinate transformation

[0101] Assume that the slant range between the SAR radar and the target is R, and the coordinates are represented by (x′, y′, z′). According to the radar downward viewing angle α, the radar coordinates are calculated as:

[0102]

[0103] The coordinates of the target model midpoint are expressed as (x″ i ,y″ i ,z″ i ) indicates that the projection coordinate of the point on the ground is (x″′ i ,y″′ i ,z″′ i ), when i is the same value, three points (x′, y′, z′), (x″ i ,y″ i ,z″ i ) and (x″′ i ,y″′ i ,z″′ i ) is on a straight line, i∈{1,2,…,K}.

[0104] For Prot For each point in the graph, its projection coordinate on the ground is calculated as (x″′ i ,y″′ i ,z″′ i ), then the shape formed by this coordinate set can represent the shadow of the target on the ground, and P proj express.

[0105]

[0106] By selecting multiple typical θ values ​​and performing coordinate rotation and coordinate transformation respectively, we can obtain the projection coordinate sets of the target on the ground at different angles, that is, we can obtain the shadow information at different angles.

[0107] In this embodiment, the shadow images with angles θ=170° and θ=269° are selected for coordinate information transformation. The P coordinate is rotated using equations (2)-(3) to obtain P rot Then, P is calculated by equations (4)-(6). rot Perform coordinate transformation to obtain the coordinate set P projected to the ground proj .

[0108] 5. Eliminate redundant target points

[0109] The redundant target point elimination in the present invention mainly includes three parts:

[0110] First, projection redundant points are eliminated: the target projection shadow obtained by coordinate transformation is matched with the shadow in the binary image to delete the information of redundant points;

[0111] Second, eliminate redundant points a priori: There is a certain downward viewing angle in the radar line of sight, which causes the target model to still have redundant structures after the redundant points are eliminated. According to the prior information of the vehicle, the vehicle body is mostly a rectangular structure, and some redundant points are eliminated based on the prior knowledge;

[0112] Third, eliminate points inside the model: After the above two steps are completed, eliminate points inside the target model and retain points on the surface of the target model.

[0113] P proj The coordinates of the midpoint are adjusted to the same scale as the coordinates of the midpoint of the image. The former is in meters and the latter is in pixels. Therefore, for P proj Perform scaling based on the resolution ρ, and convert P proj The unit is adjusted to pixels and the calculation is as follows:

[0114] P′ proj =P proj / ρ (7)

[0115] (1) Elimination of redundant projection points

[0116] Generate the first target point deletion mark variable D1, and all elements are initialized to 0. Its elements correspond one-to-one to the points in the coordinate set P of the initial target model. If the coordinates of a certain point are retained, the value of the corresponding element of the first target point deletion mark variable D1 is 0, otherwise it is 1.

[0117] First, P′ proj Coordinate centroid and I bw_s The centroids of the shadows are adjusted to be the same. Let P′ proj The coordinates of the center of mass are (x pc ,y pc ,z pc ), I bw_s The coordinates of the center of mass of the shadow are (x bwc ,y bwc ,z bwc ), P′ proj The coordinates of each point in (x bwc -x pc ,y bwc -y pc ,z bwc -z pc ) add to get the center of mass and I bw_s The same point coordinate set P′ proj2 . Traverse image I bw_s The mid-coordinate position is P′ proj2 For all pixel points, if the pixel value in the image is 0, it means that there is no shadow information at that position, then the corresponding element in the first target point deletion mark variable D1 is assigned a value of 1.

[0118] Secondly, for P′ at multiple angles θ proj , respectively perform centroid adjustment and D1 assignment, and finally obtain the final first target point deletion mark variable D1 that meets all angle conditions (assuming there are m θ, each θ corresponds to a first target point deletion mark variable D1, only when the i-th elements of the m first target point deletion mark variables D1 are all 0, the i-th element of the final mark variable D1 is assigned to 0, in other cases the i-th element is assigned to 1, i∈{1,2,…,K}).

[0119] Finally, the elements with a value of 1 in the final first target point deletion mark variable D1 correspond to the points in the coordinate set P of the initial target model, and all the coordinates of the corresponding points in the coordinate set P of the initial target model are deleted to obtain the coordinate set P′ of the target model after the projection redundant points are eliminated.

[0120] In this embodiment, the P′ at two angles θ=170° and θ=269° proj, respectively adjust the centroid and assign D1, and finally obtain the marker variable D1 that meets the conditions of each angle. Delete all the coordinates in P whose D1 value is 1, and obtain the coordinate set P′ of the target model after removing the redundant projection points. Display the midpoint position of the coordinate set P′, Figure 8 It represents the length and width of the top view after removing the redundant projection points. Fig. 9 It is the height and width front view after removing redundant projection points. Fig.10 This is the tall and long side view after redundant projection points are removed.

[0121] (2) Elimination of prior redundant points

[0122] Generate a target point deletion mark variable D2, and initialize the values ​​of K elements to 0; its elements correspond one-to-one to the points in the coordinate set P of the initial target model. If the coordinates of a certain point are retained, the value of the corresponding element of D2 is 0, otherwise it is 1.

[0123] Since most vehicles have symmetrical structures, the present invention assumes that the vehicle is symmetrical along the xOz plane, calculates the coordinates of the points symmetrical along the xOz plane in the coordinate set P of the initial target model, and finds the target points corresponding to the coordinates of the two symmetrical points, deletes the value in the second target point deletion mark variable D2, and if the values ​​are 0 and 1 respectively, the value 0 is assigned to 1.

[0124] The coordinates of the midpoints of the coordinate set P′ of the target model after removing the redundant projection points are displayed on the width and height xOz plane, as shown in Fig.11 As shown, the existence of a triangular cone at the top of the target can be seen. The cone is related to the viewing angle under the radar. The height coordinate of the bottom of the cone is calculated, and all points whose z coordinates exceed the height coordinate of the bottom of the cone in the coordinate set P′ of the target model after the projection redundant points are removed are found. The found points are matched with the elements in the second target point deletion mark variable D2, and the corresponding elements in the second target point deletion mark variable D2 are assigned a value of 1.

[0125] The coordinates of the midpoints of the coordinate set P′ of the target model after removing the redundant projection points are displayed on the length and width xOy plane, as shown in Fig.12 As shown, the existence of a triangular cone on the side of the target can be seen, and the cone is related to the angle θ. The coordinate range of the width of the target is calculated, and all points whose y coordinates exceed the coordinate range of the width of the target in the coordinate set P′ of the target model after the projection redundant points are eliminated are found. The found points are corresponded to the elements in the second target point deletion mark variable D2, and the corresponding elements in the second target point deletion mark variable D2 are assigned a value of 1.

[0126] The coordinate system where the x-axis, y-axis and z-axis are located is a three-dimensional rectangular coordinate system established with the center of mass of the target as the origin and the symmetry plane of the target as the xOz plane.

[0127] All coordinates whose corresponding second target point deletion mark variable D2 value is 1 in the coordinate set P′ of the target model after the projection redundant points are eliminated are deleted to obtain the coordinate set P″ of the target model after the prior redundant points are eliminated.

[0128] Display the midpoint position of the coordinate set P″. Fig.13 It represents the length and width of the top view after removing the prior redundant points. Fig.14 It is the height and width front view after removing the prior redundant points. Fig.15 This is the tall and long side view after removing the a priori redundant points.

[0129] (3) Elimination of internal points of the model

[0130] Generate the third target point deletion mark variable D3, initialize all elements to 1, and its elements correspond one-to-one to the points in the coordinate set P of the initial target model. If the coordinates of a certain point are retained, the corresponding element value of the third target point deletion mark variable D3 is 0, otherwise it is 1.

[0131] Construct a three-dimensional matrix T with dimensions The values ​​of its elements are all initialized to 1, and the points in the coordinate set P″ of the target model after eliminating the prior redundant points are corresponded to the elements in the three-dimensional matrix T, and the corresponding elements in the three-dimensional matrix T are assigned 0, and the correspondence is performed according to the positional relationship.

[0132] Traverse and search each element in the three-dimensional matrix T to find out all elements whose values ​​in the 3×3×3 neighborhood are all 0, correspond the found elements to the elements in the third target point deletion mark variable D3, and assign the corresponding elements in the third target point deletion mark variable D3 to 1; the correspondence is performed according to the position relationship;

[0133] The element with the value of 1 in the third target point deletion mark variable D3 is matched with the midpoint of the coordinate set P″ of the target model after the prior redundant points are eliminated, and the coordinates of the corresponding points in the coordinate set P″ of the target model after the prior redundant points are eliminated are all deleted to obtain the coordinate set P″′ of the target model after the internal points of the model are eliminated. The three-dimensional lattice structure formed by the coordinate set P″′ of the target model after the internal points of the model are eliminated is the three-dimensional reconstruction result of the target. The midpoint position of the coordinate set P″′ is displayed, as shown in FIG. Fig.16 shown.

[0134] Compared with the prior art, the present invention has the following significant advantages:

[0135] The three-dimensional structure of the target is restored using the shadow image sequence of the moving target at different angles. This invention does not rely on the imaging of the moving target or the three-dimensional imaging method. The three-dimensional structure of the target can be restored only through the shadow information of the moving target at multiple angles, providing a new idea for the recognition of moving targets in SAR images. Compared with other target three-dimensional reconstruction methods, this invention does not rely on the high-precision registration of target feature points, has a low algorithm complexity, and can achieve real-time processing.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for three-dimensional reconstruction of a target based on shadow information processing, characterized in that: Obtaining a shadow image of a target, wherein the target is a moving target, and the shadow image is a SAR image; Preprocessing the shadow image, the preprocessing including image interpolation, image binarization and image edge smoothing, to obtain a binary image after edge smoothing; Calculation of target movement direction: Calculate the target movement direction using the azimuth speed and distance speed of the moving target; Constructing a target model, wherein the target model is a three-dimensional lattice model, and the three-dimensional size of the target model is larger than its actual size; Coordinate information transformation: According to the radar attitude angle information and the target movement direction and using the coordinate transformation method, the target model is projected onto the ground to obtain two-dimensional coordinate information; the coordinate information transformation specifically includes the following steps: Calculate the coordinate rotation matrix M rot , Among them, θ is the target motion direction angle; The initial target model coordinate set P is rotated to obtain the rotated coordinate set P rot ; P rot =P×M rot Assume that the slant range between the radar and the target is R, and the coordinates are represented by (x′, y′, z′)(x1, y1, z1). According to the radar downward viewing angle α, the radar coordinates are calculated as: The coordinates of the target model midpoint are expressed as (x″ i ,y″ i ,z″ i ) indicates that the projection coordinate of the point on the ground is (x″′ i ,y″′ i ,z″′ i ), three points (x',y',z'), (x i ”,y i ”,z i ”) and (x i ”',y i ”',z i ”') in a straight line; for P rot For each point in , calculate its projection point coordinate on the ground as (z″′ i ,y″′ i ,z″′ i ), the coordinate set of the target model point projection on the ground is obtained as P proj ; Where i∈{1,2,…,K}; (x″ i ,y″ i ,z″ i ) is the rotated coordinate set P rot The coordinates of the i-th point in ; Select multiple θ values ​​and repeat the above steps to obtain the projection coordinate set of the target on the ground at different angles θ, that is, obtain the shadow information of the target on the ground at different angles θ; Redundant points of the target model are removed to obtain a three-dimensional reconstruction result of the target. The redundant point removal includes projection redundant point removal, prior redundant point removal and model internal point removal. The projection redundant point removal is to adjust the coordinate scale of the two-dimensional coordinate information obtained by coordinate information transformation and match it with the binary image after edge smoothing to delete the information of some points in the two-dimensional coordinate information.

2. The method for three-dimensional reconstruction of a target based on shadow information processing according to claim 1, characterized in that: After the coordinate information is transformed and before the redundant points are removed, the following operations are performed: The coordinate set P of the target model point projected on the ground proj Perform scaling based on the resolution ρ, and convert P proj The unit is adjusted to pixels, and the calculation formula is as follows: P′ proj =P proj / r.

3. The method for three-dimensional reconstruction of a target based on shadow information processing according to claim 2, characterized in that: The removal of redundant projection points specifically includes the following steps: Generate the first target point deletion mark variable D1, with all elements initialized to 0, and its elements correspond one-to-one to the points in the coordinate set P of the initial target model; P′ proj The centroid and edge smoothing of all coordinates in the binary image I bw_s The centroids of the middle shadows are adjusted to be the same, and the binary image I is obtained after the centroid and edge are smoothed. bw_s The same point coordinate set P′ proj2 ; Traverse the binary image I after edge smoothing bw_s The mid-coordinate position is P′ proj2 All pixels of the binary image I after edge smoothing bw_s If the pixel value of a certain pixel point is 0, the corresponding element in the first target point deletion mark variable D1 is assigned 1; For P at multiple angles θ p ' roj , respectively perform centroid adjustment and D1 assignment, and finally obtain the final first target point deletion mark variable D1 that meets the conditions of each angle; correspond the elements of the final first target point deletion mark variable D1 with a value of 1 to the midpoint of the coordinate set P of the initial target model, delete all the coordinates of the midpoint of the coordinate set P of the corresponding initial target model, and obtain the coordinate set P' of the target model after the projection redundant points are eliminated; The correspondence is performed based on the positional relationship.

4. The method for three-dimensional reconstruction of a target based on shadow information processing according to claim 3, characterized in that: P p ' roj The specific steps for adjusting the coordinate centroid are: let P′ proj The coordinates of the center of mass are (x pc ,y pc ,z pc ), binary image I after edge smoothing bw_s The coordinates of the center of mass of the shadow are (x bwc ,y bwc ,z bwc ), P′ proj The coordinates of each point in (x bwc -x pc ,y bwc -y pc ,z bwc -z pc ) added.

5. The method for three-dimensional reconstruction of a target based on shadow information processing according to claim 3, characterized in that: The priori redundant point elimination specifically includes the following steps: Generate a second target point deletion mark variable D2, whose elements correspond one-to-one to the points in the coordinate set P of the initial target model, and the values ​​of K elements are initialized to 0; Assuming that the target is symmetrical along the xOz plane, calculate the coordinates of the points symmetrical along the xOz plane in the coordinate set P of the initial target model, and find the target points corresponding to the coordinates of the two symmetrical points. Delete the value in the second target point deletion flag variable D2. If the values ​​are 0 and 1 respectively, assign the value 0 to 1; Display the midpoint coordinates of the coordinate set P' of the target model after the redundant projection points are removed on the width and height xOz plane, find the target top triangular pyramid, calculate the height coordinates of the bottom of the target top triangular pyramid, find all the points whose z coordinates in the coordinate set P' of the target model after the redundant projection points are removed exceed the height coordinates of the bottom of the target top triangular pyramid, correspond the found points to the elements in the second target point deletion mark variable D2, and assign the corresponding elements in the second target point deletion mark variable D2 to 1; Calculate the coordinate range of the width of the target, find all points whose y coordinates exceed the coordinate range of the width of the target in the coordinate set P' of the target model after the projection redundant points are removed, correspond the found points to the elements in the second target point deletion mark variable D2, and assign the corresponding elements in the second target point deletion mark variable D2 to 1; Delete all the coordinates of the second target point deletion flag variable D2 with a value of 1 in the coordinate set P' of the target model after the projection redundant points are eliminated, and obtain the coordinate set P" of the target model after the prior redundant points are eliminated; Among them, the coordinate system in which the coordinate set P″ of the target model after the prior redundant points are eliminated is a three-dimensional rectangular coordinate system with the center of mass of the target as the origin and the symmetry plane of the target as the xOz plane. The x-axis, y-axis and z-axis of the three-dimensional rectangular coordinate system correspond to the length, width and height directions of the target respectively; the correspondence is performed according to the positional relationship.

6. The method for three-dimensional reconstruction of a target based on shadow information processing according to claim 5, characterized in that: The model internal point elimination specifically includes the following steps: Generate a third target point deletion flag variable D3, initialize all elements to 1, and have a one-to-one correspondence with the points in the coordinate set P of the initial target model; Construct a three-dimensional matrix T whose size is The values ​​of its elements are all initialized to 1, and the points in the coordinate set P″ of the target model after eliminating the prior redundant points are matched with the elements in the three-dimensional matrix T, and the corresponding elements in the three-dimensional matrix T are assigned 0, and the matching is performed according to the position relationship; Traverse and search each element in the three-dimensional matrix T to find out all elements whose values ​​in the 3×3×3 neighborhood are all 0, correspond the found elements to the elements in the third target point deletion mark variable D3, and assign the corresponding elements in the third target point deletion mark variable D3 to 1; the correspondence is performed according to the position relationship; The element with the element value of 1 in the third target point deletion mark variable D3 corresponds to the midpoint of the coordinate set P″ of the target model after the prior redundant points are eliminated, and the coordinates of all the corresponding points in the coordinate set P″ of the target model after the prior redundant points are eliminated are deleted to obtain the coordinate set P″′ of the target model after the internal points of the model are eliminated. The three-dimensional lattice structure formed by the coordinate set P″′ of the target model after the internal points of the model are eliminated is the three-dimensional reconstruction result of the target.

7. The method for three-dimensional reconstruction of a target based on shadow information processing according to claim 6, characterized in that: The target motion direction calculation specifically includes the following steps: Assume that the direction of movement is replaced by angle information, with the radar line of sight as 0° and rotating clockwise, the target movement direction angle range is [0°, 360°); The velocity information is smoothed using the neighborhood mean method, and the target's moving direction angle θ is calculated as: Among them, V r and V a They are the range velocity and azimuth velocity of the target after smoothing.

8. The method for three-dimensional reconstruction of a target based on shadow information processing according to claim 7, characterized in that: The shadow image preprocessing specifically includes: Assume that the shadow image is represented by I. According to the radar viewing angle, the shadow image I is re-interpolated so that the distance information of the image is the distance information of the actual scene, and the image I is obtained. ins ; Set the binarization threshold T bw , for image I ins After binarization, the erosion and expansion operations are performed to reduce the roughness of the shadow edge. Finally, the largest connected domain in the image is retained as the shadow area to obtain the binary image I bw ; Extract the shadow edge in the binary image, smooth the shadow edge through smoothing operation, and then fill the inside of the edge to obtain the binary image I after edge smoothing bw_s .

Images