A three-dimensional reconstruction method and system based on two-dimensional forward-looking sonar images

Abstract

The application discloses a kind of three-dimensional reconstruction method and system based on two-dimensional forward-looking sonar image, the method includes: step one: for the two-dimensional forward-looking sonar sequence collected, introduce based on mahalanobis distance detection method, determine the initial profile of local binary fitting level set segmentation method, further, using a local binary fitting (LBF) level set segmentation method, determine the target to be reconstructed region based on the initial profile detected;Step two: using the registration method based on phase correlation, the target region determined in step one is registered, and the relative displacement of sonar movement is calculated.Step three: based on the results of the above steps, using blind deconvolution algorithm, reconstructs the three-dimensional point cloud image of target.Compared with the prior art, the three-dimensional reconstruction of the target can be performed without pre-acquiring the horizontal position information of the sonar, and has certain anti-noise ability.The method disclosed in the application is more suitable for three-dimensional reconstruction of underwater targets.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a three-dimensional reconstruction method and system based on two-dimensional forward-looking sonar images. Background Technology

[0002] In recent years, with the continuous development of the ocean, the estimation of the shape and position of underwater objects has become a hot topic. Applications such as detecting pipeline leaks and detecting and identifying potential threats on the seabed all require accurate estimation of the position and shape information of underwater objects.

[0003] Although forward-looking sonar has a large horizontal aperture, enabling imaging of a wide area in water, it cannot accurately obtain the elevation angle information of the target due to ambiguity in the vertical aperture direction, thus preventing 3D imaging of underwater targets. Existing 2D forward-looking sonar 3D reconstruction methods require precise control of the sonar trajectory to ensure that the positions of all points of the target to be reconstructed remain consistent in the sonar image.

[0004] Therefore, there is an urgent need for a target reconstruction method based on two-dimensional forward-looking sonar images to achieve three-dimensional reconstruction of underwater targets. Summary of the Invention

[0005] In order to solve the technical problems existing in the background art, the present invention aims to provide a three-dimensional reconstruction method and system based on two-dimensional forward-looking sonar images, so as to realize the recovery of the elevation angle of the target using two-dimensional forward-looking sonar images.

[0006] To solve the technical problem, the technical solution of the present invention is as follows:

[0007] A method for three-dimensional reconstruction based on two-dimensional forward-looking sonar images, the method comprising:

[0008] S1: For the collected two-dimensional forward-looking sonar sequence, the initial contour in the image is determined by the Mahalanobis distance detection method. Based on the detected initial contour, the target region is accurately segmented by the Local Binary Fitting (LBF) level set segmentation method. The target region to be reconstructed is determined by the segmentation results.

[0009] S2: For the target area determined in step S1, a certain frame is selected as the reference frame by a phase-correlation-based registration method, and other sonar image frames are regarded as motion frames. Registration is performed using a formula, and the relative displacement of the sonar is calculated.

[0010] S3: Based on the target area obtained in step S1 and the relative displacement of the sonar obtained in step S2, a blind deconvolution algorithm is used to reconstruct the three-dimensional point cloud image of the target.

[0011] Furthermore, in step S1, a Mahalanobis distance-based detection method is introduced to determine the initial contour of the local binary fitting level set segmentation method. This method detects points in the image using inner and outer sliding windows to determine whether they are target regions. The form of its test statistic is as follows:

[0012]

[0013] In the formula, D(x) is the test statistic of the algorithm, U0 represents the case where there is no target in the region, and event U1 represents the case where there is a target in the region. s u is the pixel mean of the inner window. s The average pixel value of the outer window. Let V be the variance of the exterior window;

[0014] The algorithm estimates the initial contour of a target in a sonar image by comparing the calculated test statistic D(x) with a predefined threshold.

[0015] Further, in step S1: a local binary fitting LBF level set segmentation method is used, based on the detected initial contour of the target, to determine the target's reconstructed region using the initial contour determined by this algorithm. The energy functional form of this method is:

[0016]

[0017] In the formula, F(φ,f1,f2) is the energy functional of the level set, and ε LBF (φ,f1,f2) represents the area term of the local binary fitting method. As a penalty item, The length term, μ, and v are the coefficients of the area term and the energy term, respectively;

[0018] ε LBF The specific form of (φ,f1,f2) is:

[0019] ε LBF (φ,f1,f2)=λ1∫[∫K σ (xy)|I(y)-f1(x)| 2 H(φ(y))dy]dx+λ2∫[∫K σ (xy)|I(y)-f2(x)| 2 ·(1-H(φ(y)))dy]dx

[0020] In the formula, λ1 and λ2 are constants, K is the kernel function, I is the sonar image frame, H is the Heaviside function, φ is the level set function, and f1(x) and f2(x) are the values ​​near the fitting point x, which change as the position of the fitting point changes.

[0021] The specific form is as follows:

[0022]

[0023] In the formula, Ω represents the image domain. Gradient symbol;

[0024] The specific form is

[0025]

[0026] In the formula, δ is the derivative of the Heaviside function;

[0027] The algorithm minimizes the energy functional using gradient descent to iterate over the target contour, and its formula is as follows:

[0028]

[0029] In the formula, e1 and e2 have the following forms:

[0030] e1(x)=∫ Ω K(yx)|I(x)-f1(y)| 2 dy

[0031] e2(x)=∫ Ω K(yx)|I(x)-f2(y)| 2 dy

[0032] Based on the discretization form of the above formula, by setting the time step, a refined outline of the target can be obtained.

[0033] Furthermore, in step S2:

[0034] Using a phase-correlation-based registration method, the image segmentation result (target region) determined in step S1 is registered. A certain frame is selected as the reference frame, and other sonar image frames are regarded as moving frames and registered to the reference frame. The relative displacement of the sonar motion is calculated. The registration formula is as follows:

[0035]

[0036] In the formula, F1(ξ,η) is the Fourier transform result of the reference frame image, F2(ξ,η) is the Fourier transform result of the motion frame image, and x0 and y0 are the relative displacements of the sonar motion.

[0037] The formula for the cross-power spectrum of the reference frame and the moving frame is as follows:

[0038]

[0039] In the formula, H(ξ,η) is the cross power spectrum, and * represents complex conjugate;

[0040] By performing an inverse Fourier transform on the cross-power spectrum of the above equation, the relative displacement of the sonar can be estimated by finding the peak value of the obtained pulse function.

[0041] Furthermore, step S3 specifically includes:

[0042]

[0043] In the formula: I rθ (z) represents the grayscale value of the sonar image r,θ at height z after registration, β r V is the vertical directivity index corresponding to point r in the sonar image. rθ Let ρ(z,u,n) be the coefficient of the existence term of the target, which is the point cloud image to be solved. Let ρ(z,u,n) be the cosine of the angle between the incident sound wave and the normal vector n of the object surface. Using the discretized form of the above formula, the three-dimensional point cloud of the target is reconstructed by solving the non-negative least squares solution.

[0044] Furthermore, in step S3, the discrete form of the formula for the grayscale value of the sonar image r,θ at the registered height z is shown below:

[0045]

[0046] In the formula, z i It is the z-coordinate of the i-th frame image sequence, and P is the total number of points in the vertical direction of the point cloud to be reconstructed;

[0047] The above formula can be rearranged into the following form:

[0048] I rθ =A N×P ·b rθ

[0049] In the formula b rθ =V krθ ;

[0050] Solve the equation using the nonnegative least squares method to obtain b. rθ The value of is the point cloud after the target is reconstructed in 3D.

[0051] A three-dimensional reconstruction system based on two-dimensional forward-looking sonar images, the system being used to perform any of the methods described above, the system comprising:

[0052] Target domain segmentation module: For acquiring sonar image sequences, a fast detection method based on Mahalanobis distance is used to determine the approximate outline of the target, and then the local binary level set method is used to accurately determine the target outline;

[0053] Registration module: Uses phase correlation algorithm to register the target contour extracted by image segmentation module and estimate inter-frame displacement;

[0054] 3D Reconstruction Module: Using the precise contours of the target segmentation module and the inter-frame displacements calculated by the registration module, the target acquired by the forward-looking sonar image sequence is reconstructed in 3D.

[0055] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a three-dimensional reconstruction method based on two-dimensional forward-looking sonar images as described above.

[0056] A computer-readable storage medium storing a computer program that, when executed by a processor, implements a three-dimensional reconstruction method based on two-dimensional forward-looking sonar images as described above.

[0057] Compared with the prior art, the advantages of the present invention are as follows:

[0058] Compared with traditional methods, this invention eliminates the need to pre-obtain the X and Y displacements of the sonar; it only requires the Z-axis relative displacement of the sonar to perform 3D target reconstruction. Furthermore, because the target position and contour are pre-segmented, background interference can be eliminated, providing a certain degree of noise resistance. The method described in this invention is more suitable for 3D reconstruction of underwater targets. Attached Figure Description

[0059] Figure 1 This is a flowchart of the three-dimensional reconstruction method based on two-dimensional forward-looking sonar images described in this invention;

[0060] Figure 2 This is the result of the three-dimensional reconstruction method based on two-dimensional forward-looking sonar images described in this invention. Detailed Implementation

[0061] The specific implementation of the present invention is described below with reference to embodiments:

[0062] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0063] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0064] Example 1:

[0065] Combination Figure 1 This invention proposes a three-dimensional reconstruction method based on two-dimensional forward-looking sonar images, the method comprising:

[0066] Step 1: This invention introduces a Mahalanobis distance-based detection method to determine the initial contour of a local binary fitting level set segmentation method. This method uses inner and outer sliding windows to detect points in the image to determine whether they are target regions. The form of its test statistic is as follows:

[0067]

[0068] In the formula, D(x) is the test statistic of the algorithm, U0 represents the case where there is no target in the region, and event U1 represents the case where there is a target in the region. s u is the pixel mean of the inner window. b The average pixel value of the outer window. Let V be the variance of the exterior window.

[0069] The algorithm estimates the initial contour of a target in a sonar image by comparing the calculated test statistic D(x) with a predefined threshold.

[0070] Furthermore, in step one, a local binary fitting (LBF) level set segmentation method is used to determine the target's reconstructable region based on the detected initial target contour. The energy functional form of this method is as follows:

[0071]

[0072] In the formula, F(φ,f1,f2) is the energy functional of the level set, and ε LBF (φ,f1,f2) represents the area term of the local binary fitting method. As a penalty item, The length term, μ, and υ are the coefficients of the area term and the energy term, respectively.

[0073] ε LBF The specific form of (φ,f1,f2) is:

[0074] ε LBF (φ,f1,f2)=λ1∫[∫K σ(xy)|I(y)-f1(x)| 2 H(φ(y))dy]dx+λ2∫[∫K σ (xy)|I(y)-f2(x)| 2 ·(1-H(φ(y)))dy]dx

[0075] In the formula, λ1 and λ2 are constants, K is the kernel function, I is the sonar image frame, H is the Heaviside function, φ is the level set function, and f1(x) and f2(x) are values ​​near the fitting point x, which change as the position of the fitting point changes.

[0076] The specific form is as follows:

[0077]

[0078] In the formula, Ω represents the image domain. Gradient symbol;

[0079] The specific form is

[0080]

[0081] In the formula, δ is the derivative of the Heaviside function.

[0082] The algorithm minimizes the energy functional using gradient descent to iterate over the target contour, and its formula is as follows:

[0083]

[0084] In the formula, e1 and e2 are in the forms of respectively

[0085] e1(x)=∫ Ω K(yx)|I(x)-f1(y)| 2 dy

[0086] e2(x)=∫ Ω K(yx)|I(x)-f2(y)| 2 dy

[0087] Based on the discretization form of the above formula, by setting the time step, a refined outline of the target can be obtained.

[0088] Step 2: Using a phase-correlation-based registration method, register the target region determined in Step 1. Select a frame as the reference frame, treat other sonar image frames as moving frames, and register them to the reference frame. Calculate the relative displacement of the sonar motion. In the case of only rigid translation, the relationship between the moving frame and the reference frame is as follows:

[0089] f2(x,y)=f1(x-x0,y-y0)

[0090] In the formula, f1(x,y) and f2(x,y) are the reference frame and the motion frame, respectively, and x0 and y0 are the relative displacements of the x and y axes.

[0091] Furthermore, its Fourier transform form is:

[0092]

[0093] Furthermore, the formula for the cross-power spectrum of the reference frame and the moving frame is as follows:

[0094]

[0095] In the formula, H(ξ,η) is the cross power spectrum, and * represents complex conjugate.

[0096] By performing an inverse Fourier transform on the cross-power spectrum of the above equation, the relative displacement of the sonar can be estimated by finding the peak value of the obtained pulse function.

[0097] Step 3: Based on the sonar image sequence segmentation results obtained in Step 1 and the sonar relative displacement obtained in Step 2, a blind deconvolution algorithm is used to reconstruct the 3D point cloud image of the target. The formula is as follows:

[0098]

[0099] In the formula I rθ (z) represents the grayscale value of the sonar image r,θ at height z after registration, β r V is the vertical directivity index corresponding to point r in the sonar image. rθ The coefficients of the existence term of the objective are the point cloud image to be solved, and ρ(z,u,n) is the cosine of the angle between the incident sound wave and the normal vector n of the object surface.

[0100] Furthermore, the discrete form of the above equation is shown below:

[0101]

[0102] In the formula, z i It is the z-coordinate of the i-th frame image sequence, and P is the total number of points in the vertical direction of the point cloud to be reconstructed.

[0103] Furthermore, the above formula can be rearranged into the following form.

[0104] I rθ =A N×P ·b rθ

[0105] In the formula b rθ =V krθ.

[0106] Furthermore, this invention uses the nonnegative least squares method to solve the equation, obtaining b. rθ The value of is the point cloud after the target is reconstructed in 3D.

[0107] Figure 2 The left side shows one of the frames in the sonar image sequence, and the right side shows the point cloud reconstruction result. This is the result of processing the simulated sonar image sequence using the algorithm.

[0108] Example 2:

[0109] This invention proposes a 3D reconstruction system based on forward-looking sonar images. This system can be used to implement the aforementioned 3D reconstruction method based on forward-looking sonar images. Specifically, the system includes:

[0110] Target domain segmentation module: For sonar image sequences, a fast detection method based on Mahalanobis distance is used to determine the approximate outline of the target, and then the local binary level set method is used to accurately determine the target outline;

[0111] Registration module: Using the phase correlation algorithm described above, the target contours extracted by the image segmentation module are registered, and the inter-frame displacement is estimated;

[0112] 3D Reconstruction Module: Using the target location information from the segmentation module and the inter-frame displacement calculated by the registration module, the module performs 3D reconstruction of the target acquired from the forward-looking sonar image sequence.

[0113] Example 3:

[0114] This embodiment provides a terminal device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment can be used in the operation of a three-dimensional reconstruction method based on forward-looking sonar images, including the following steps:

[0115] Step 1: For the acquired two-dimensional forward-looking sonar sequence, since the traditional local binary fitting level set segmentation method is sensitive to the initial contour, an incorrect initial contour will affect the final contour segmentation result of the target. This invention introduces a Mahalanobis distance-based detection method to determine the initial contour of the local binary fitting level set segmentation method. This method detects each point in the image by using inner and outer sliding windows.

[0116] Step 2: Using a phase-correlation-based registration method, register the target region determined in Step 1. Select a frame as the reference frame, treat other sonar image frames as moving frames, and register them to the reference frame. Calculate the relative displacement of the sonar motion. The registration formula is as follows:

[0117]

[0118] In the formula, F1(ξ,η) is the Fourier transform result of the reference frame image, F2(ξ,η) is the Fourier transform result of the motion frame image, and x0 and y0 are the relative displacements of the sonar motion.

[0119] Step 3: Based on the segmentation results obtained in Step 1 and the relative sonar displacement obtained in Step 2, the 3D point cloud image of the target is reconstructed using a blind deconvolution algorithm, and its formula is as follows.

[0120]

[0121] In the formula I rθ(z) represents the grayscale value of the sonar image r,θ at height z after registration, β r V is the vertical directivity index corresponding to point r in the sonar image. rθ The coefficients of the existence term of the objective are the point cloud image to be solved, and ρ(z,u,n) is the cosine of the angle between the incident sound wave and the normal vector n of the object surface.

[0122] Example 4:

[0123] This embodiment provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0124] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the three-dimensional reconstruction method based on forward-looking sonar images in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps:

[0125] Step 1: For the acquired two-dimensional forward-looking sonar sequence, since the traditional local binary fitting level set segmentation method is sensitive to the initial contour, an incorrect initial contour will affect the final contour segmentation result of the target. This invention introduces a Mahalanobis distance-based detection method to determine the initial contour of the local binary fitting level set segmentation method. This method detects each point in the image by using inner and outer sliding windows.

[0126] Step 2: Using a phase-correlation-based registration method, register the target region determined in Step 1. Select a frame as the reference frame, treat other sonar image frames as moving frames, and register them to the reference frame. Calculate the relative displacement of the sonar motion. The registration formula is as follows:

[0127]

[0128] In the formula, F1(ξ,η) is the Fourier transform result of the reference frame image, F2(ξ,η) is the Fourier transform result of the motion frame image, and x0 and y0 are the relative displacements of the sonar motion.

[0129] Step 3: Based on the segmentation results obtained in Step 1 and the relative sonar displacement obtained in Step 2, the 3D point cloud image of the target is reconstructed using a blind deconvolution algorithm, and its formula is as follows.

[0130]

[0131] In the formula I rθ (z) represents the grayscale value of the sonar image r,θ at height z after registration, β r V is the vertical directivity index corresponding to point r in the sonar image. rθ Let ρ(z,u,n) be the coefficient of the existence term of the objective, which is the point cloud image to be solved, and let ρ(z,u,n) be the cosine of the angle between the incident sound wave and the normal vector m of the object surface.

[0132] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0134] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0135] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0136] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0137] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A method for 3D reconstruction based on 2D forward-looking sonar images, characterized in that, The method comprises: S1: determining the initial contour in the image by using a Mahalanobis distance detection method, and accurately segmenting the target region by using a local binary fitting (LBF) level set segmentation method based on the detected initial contour, and determining the target region to be reconstructed through the segmentation result; S2: using a phase correlation-based registration method to register the target region determined in step one, selecting a certain frame as a reference frame, regarding other sonar image frames as motion frames, registering the motion frames to the reference frame, calculating the relative displacement of the sonar motion, and in the case of only existing rigid translation, the relationship between the motion frame and the reference frame is as follows: ， wherein , are a reference frame and a motion frame, respectively, , are the relative displacements of the x, y axes; The Fourier transform form is: ， The formula form of the cross power spectrum of the reference frame and the motion frame is: ， wherein is the cross power spectrum, * is the complex conjugate; The inverse Fourier transform of the cross power spectrum of the above formula is obtained, and the relative displacement of the sonar can be estimated by finding the peak value of the obtained impulse function; S3: using a blind deconvolution algorithm to reconstruct the three-dimensional point cloud image of the target based on the target region obtained in step S1 and the relative displacement of the sonar obtained in step S2, and the formula is: ， In the formula, For the height after registration sonar images grayscale value, For sonar images The corresponding vertical directivity index at that location, The coefficients of the existence term of the objective are the point cloud image to be solved. The incident sound wave and the normal vector of the object surface The cosine value of the included angle; the blind deconvolution algorithm constructed using this formula reconstructs the three-dimensional point cloud image of the target.

2. The method according to claim 1, wherein, In the step S1, the initial contour of the local binary fitting level set segmentation method is determined by introducing the Mahalanobis distance detection method, which detects each point in the image by using the inner and outer sliding window to determine whether it is a target region, and the expression form of the test statistic is: ， wherein is the test statistic for the algorithm, represents the case where the region is free of targets, event represents the case where the region is free of targets, event is the mean of the pixels of the inner window, is the mean of the pixels of the outer window, is the variance of the outer window; The algorithm estimates the initial contour of the target in the sonar image by calculating the test statistic The initial contour of the target in the sonar image is estimated by discriminating with the set threshold value.

3. The method of claim 1, wherein, In the step S1: using a local binary fitting (LBF) level set segmentation method to determine the target region to be reconstructed based on the detected initial contour of the target, using the initial contour determined based on the algorithm, and the energy functional form of the method is: ； wherein is an energy functional of a level set, is an area term of a local binary fitting method, is a penalty term, is a length term, , are coefficients of the penalty term and the length term, respectively. In particular, the application relates to a compound of formula (I) ， wherein , is a constant, is a kernel function, is a sonar image frame, is a Heaviside function, is a level set function, , is a value near the fitting point changes as the position of the fitting point changes; In particular, the application relates to a compound of formula (I) ， wherein is the image domain, is the gradient sign; In particular, the compound of formula (I) is: ， wherein is the derivative of the Heaviside function; The algorithm minimizes the energy functional by using the gradient descent method to realize the iteration of the target contour, and the formula form is as follows: ， wherein , in the form of ， ， wherein , is the value near the fitting point ; according to the discrete form of the above equation, the time step is set, and finally the refined profile of the target can be obtained.

4. The method of claim 1, wherein, In the step S3, the registered height of the sonar image The discrete form of the formula for the gray value is as follows: ， In the formula, is the z coordinate of the i-th image sequence, and P is the total number of points in the vertical direction of the point cloud to be reconstructed. The simplified form of the above formula is: ， In the formulae , ; Using non-negative least square method to solve the equation, the value of is obtained, that is, the point cloud after the target three-dimensional reconstruction.

5. A 3D reconstruction system based on 2D forward-looking sonar images, characterized in that, The system is used to execute the method in any one of claims 1-4, and the system comprises: A target domain segmentation module: for acquiring a sonar image sequence, using a fast detection method based on Mahalanobis distance to determine the contour of the target, and then using a local binary fitting (LBF) level set method to accurately determine the target contour; A registration module: using a phase correlation algorithm to register the target contour extracted by the image segmentation module, and estimating the inter-frame displacement; A three-dimensional reconstruction module: using the accurate contour of the target domain segmentation module and the inter-frame displacement calculated by the registration module to perform three-dimensional reconstruction on the target acquired by the forward-looking sonar image sequence.

6. A computer device, comprising: A computer readable storage medium stores a computer program, and the program is executed on a processor to realize the three-dimensional reconstruction method based on two-dimensional forward-looking sonar images in any one of claims 1 to 4.

7. A computer readable storage medium characterized in that, The computer readable storage medium stores a computer program, and the program is executed on a processor to realize the three-dimensional reconstruction method based on two-dimensional forward-looking sonar images in any one of claims 1 to 4.

Images