Extensible surround dual camera array fragment kill field three-dimensional reconstruction damage assessment system

Abstract

The application discloses an extensible surrounding dual-camera array fragment damage field three-dimensional reconstruction damage evaluation system, which comprises at least two structural members, the structural members comprising a shell; at least two high-speed cameras rigidly mounted in the shell; an inertial measurement unit arranged in the shell; a worm-gear hinge structure connected between the optical axes of the two high-speed cameras, the worm-gear hinge structure enabling the included angle θ between the two optical axes to be continuously adjustable within the range of 15°-150°; and a set of three-dimensional dynamic reconstruction and damage evaluation software, wherein the system is capable of quickly completing field layout and overall camera external parameter calibration, measuring the speed, size, trajectory and density of fragments in all directions, generating three-dimensional damage field point clouds by one key, and driving a damage model to perform interactive animation demonstration and quantitative evaluation.

Description

Technical Field

[0001] This invention relates to the technical field of ammunition terminal damage testing, specifically to a scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system. Background Technology

[0002] Terminal damage testing of ammunition is a crucial step in military scientific research and weapons development. Existing high-speed photography methods typically employ single-station or dual-station two-dimensional measurements. Due to the field of view angle being less than 30°, it is difficult to obtain the three-dimensional coordinates and velocity vectors of omnidirectional fragments, resulting in significant discrepancies between the calculated results of key indicators such as kill radius and kill area and actual live-fire statistics.

[0003] When multiple cameras are arranged in a ring for measurement, each camera needs to be calibrated individually. On-site setup and calibration of the target usually takes more than 2 hours. Furthermore, the shock wave from the explosion can easily cause the target position to shift and the external parameters to drift, which in turn can lead to splicing misalignment and fragment density counting errors.

[0004] Traditional post-processing software primarily outputs static tables and cannot generate interactive 3D animation models. This makes it difficult for assessors to intuitively judge the differences in damage from different directions and distances, and it cannot meet the output requirements for comprehensive indicators such as pre-fabricated / natural fragment size, velocity decay curves, trajectories, and densities in one go.

[0005] Furthermore, existing fragment velocity range, accuracy, and recognition rate specifications generally fail to meet the requirements of military testing guidelines. In areas with sparse fragmentation, false positives or false negatives often lead to discontinuous kill fields, affecting the accuracy of the assessment.

[0006] To address the aforementioned issues, we offer a scalable surround dual-camera array fragment kill field 3D reconstruction damage assessment system to resolve these problems. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a scalable surround dual-camera array fragment kill field 3D reconstruction damage assessment system. Through a three-in-one solution of "structural component level - array level - software level", it enables rapid on-site deployment and calibration of all camera extrinsic parameters, achieves 360° omnidirectional fragment velocity, size, trajectory, and density measurement, generates 3D kill field point cloud with one click, and drives the damage model to perform interactive animation demonstration and quantitative assessment.

[0008] To achieve the above objectives, the present invention employs a scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system, comprising: At least two structural components, the structural components including: case; At least two high-speed cameras are located inside the housing and rigidly mounted on a common substrate; An inertial measurement unit located inside its housing; A dual-mode receiver for a global navigation satellite system is disposed inside the housing; A worm gear-worm hinge structure is connected between the optical axes of the two high-speed cameras, which allows the included angle θ between the two optical axes to be continuously adjustable within the range of 15° to 150°; V-shaped quick-release guide rail is provided at the bottom of the structural component; The system also includes a set of three-dimensional dynamic reconstruction and damage assessment software.

[0009] As a further optimization of the above scheme, the system is implemented through two-level rapid extrinsic parameter calibration, the two-level rapid extrinsic parameter calibration including: Internal calibration: Before leaving the factory, the relative pose between the high-speed camera and the inertial measurement unit inside each structural component is calibrated to establish a fixed three-dimensional baseline and camera-inertial measurement unit extrinsic parameters; Array-level calibration: At least one structural component is arranged in an equidistant array around the center of the field. A miniature handheld ball with LEDs is thrown at the center of the field. The external attitude of all structural components is calculated by combining the ball center images synchronously acquired by the cameras in each structural component and the absolute attitude output by the inertial measurement unit.

[0010] As a further optimization of the above scheme, the joint adjustment in the array-level calibration is achieved by minimizing a cost function, which includes: ; in, This is the visual reprojection error term. For the absolute attitude constraint terms of the inertial measurement unit, These are prior constraints.

[0011] As a further optimization of the above scheme, the visual reprojection tail error term Defined as: ; in, Let be the pixel coordinates of the center of the sphere of the j-th camera in the k-th frame of the i-th structural component. Let J be the intrinsic parameter matrix of camera j. , World coordinate system to camera The transformation matrix, The pose of the structural component to be optimized. The known quantities are those specified by the manufacturer. Let be the three-dimensional coordinates of the center of the sphere in the k-th frame of the world coordinate system. This is a camera projection model.

[0012] As a further optimization of the above scheme, the absolute attitude constraint term of the inertial measurement unit... Defined as ; Where R_world2imu_i_measured is the absolute rotation matrix directly measured by the inertial measurement unit, R_world2imu_i_optimized is the rotation matrix part of the optimization variables, and Log(...) is a vector that maps the difference of the rotation matrix to the Lie algebra and calculates its norm.

[0013] As a further optimization of the above solution, the three-dimensional dynamic reconstruction and damage software includes: A module for preprocessing high-speed sequence images acquired by the high-speed camera; The weighted index fragment extraction module is used to detect single-frame fragments in real time from the preprocessed image, output the three-dimensional coordinates of the fragments, obtain the trajectory, calculate the speed and size, and perform density statistics; The three-dimensional kill field reconstruction module is used to import the point cloud with position, velocity, size and trajectory attributes output by the full index fragment extraction module into the three-dimensional engine and generate a scattering animation. The damage quantification assessment module performs fragment-target collision detection, calculates the number of hits and penetration probability based on the three-dimensional model generated by the three-dimensional kill field reconstruction module, and outputs the kill radius, kill area and expected damage probability.

[0014] As a further optimization of the above scheme, in the full-index fragment extraction module, the lightweight YOLO-FragNet algorithm is used for real-time detection of single-frame fragments, and the three-dimensional coordinates of the output fragments are obtained by stereo matching and triangulation, combined with temporal Kalman filtering to obtain the trajectory.

[0015] As a further optimization of the above solution, the three-dimensional kill field reconstruction module imports point clouds with "position + velocity + size + trajectory" attributes into the Unity3D / Unreal engine, and uses a GPU particle system to drive each fragment according to the real velocity vector to generate a 0-5s scattering animation.

[0016] As a further optimization of the above solution, the damage quantification assessment module has a built-in vulnerability database for personnel, vehicles, radar, etc.

[0017] As a further optimization of the above solution, the bottom V-shaped quick-release guide rail of the structural component can be snapped onto a ground tripod, a vertical mast, or a circular guide rail.

[0018] The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system of the present invention has the following beneficial effects: 1. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system of the present invention effectively reduces the total time of on-site deployment and calibration through two-level calibration of "structural components-array", greatly improving testing efficiency. At the same time, it realizes 360° surround measurement, solves the problem of small field of view of traditional solutions, realizes omnidirectional fragment kill area calculation without blind spots, and improves the accuracy of damage assessment. 2. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system of the present invention can generate interactive three-dimensional animation with one click. The assessors can replay the animation from any perspective in the virtual scene, providing an intuitive basis for warhead optimization and protection design, and improving assessment efficiency and decision quality. 3. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system of the present invention is based on the GPU parallel + YOLO-FragNet fragment recognition and high-precision velocity measurement integrated algorithm, as well as a comprehensive density statistics method, which improves the accuracy and reliability of fragment extraction and effectively solves the problem of discontinuous kill field caused by false detection and missed detection in sparse areas. 4. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system of the present invention, with "sphere center-IMU" joint adjustment algorithm, effectively reduces the sphere center three-dimensional reconstruction error, reduces array splicing misalignment, and ensures that the entire camera array forms a highly consistent global measurement network.

[0019] Specific embodiments of the present invention are disclosed in detail with reference to the following description and accompanying drawings, indicating how the principles of the present invention can be adopted. It should be understood that the embodiments of the present invention are not limited in scope as a result, and the embodiments of the present invention include many changes, modifications and equivalents. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the axial structure of the structural component in this invention; Figure 2 This is a schematic diagram of the surrounding array arrangement of the structural components of the present invention; Figure 3 This is a schematic diagram of the three-dimensional trajectory of the three-dimensional kill field in this invention; Figure 4 This is a schematic diagram of the statistical results of the three-dimensional kill field in this invention; Figure 5 This is a schematic diagram of the animation demonstration interface (which can be paused / sliced) in this invention.

[0021] In the diagram: 1. Housing; 2. High-speed camera; 3. Inertial measurement unit; 4. Worm gear-worm hinge structure. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, it should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of the invention.

[0023] It should be noted that when an element is referred to as "set on" or "provided with" another element, it can be directly on the other element or there may be an intermediate element. When an element is referred to as "connected to" or "connected to" another element, it can be directly connected to the other element or there may be an intermediate element at the same time. "Fixed connection" means fixed connection. There are many ways of fixed connection, which are not within the scope of protection of this document. The terms "vertical", "horizontal", "left", "right" and similar expressions used in this document are only for illustrative purposes and do not represent the only implementation method.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in the specification herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Please refer to the instruction manual appendix. Figure 1-5 The present invention provides a first embodiment of a scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system, the specific details of which are as follows: The hardware of the scalable surround dual-camera array fragment kill field 3D reconstruction damage assessment system in this embodiment includes the following components: Structural components: The system includes at least two structural components. For example, in this embodiment, there are 6 structural components, and each structural component adopts a two-piece anti-seismic titanium alloy shell 1.

[0025] High-speed camera 2: Inside each housing 1, two high-speed cameras 2 are mounted on a rigid common substrate. The frame rate of the high-speed cameras 2 is ≥25000fps. In some embodiments, the high-speed camera 2 is a Phantom v2512. These two high-speed cameras 2 face outward and are used to capture fragment motion.

[0026] Inertial Measurement Unit 3 and BeiDou GNSS Receiver: Each housing 1 also integrates an Inertial Measurement Unit 3 (IMU) and a BeiDou GNSS receiver, which are used to record the attitude of the structural components and accurate global positioning information in real time.

[0027] Hinge: Connecting the optical axes of the two high-speed cameras 2 is a worm gear hinge structure 4. This hinge features electronic calibration feedback, allowing the included angle θ between the two optical axes to be precisely adjusted within a range of 15° to 150°, for example, with an adjustment accuracy of ±0.05°. This allows the structure to adapt to different testing scenarios and measurement requirements.

[0028] V-shaped quick-release rails: Each structural component is designed with a V-shaped quick-release rail at the bottom.

[0029] Synchronization Box: The unified synchronization box connects to all structural components via a data cable, providing a high-precision hardware trigger signal to ensure that all cameras start shooting at the same time.

[0030] Protection: Each high-speed camera 2 is equipped with a protective window at the front end, which can withstand overpressure shock waves and effectively resist the impact and fragmentation generated by the explosion.

[0031] Specifically, during field deployment, in a typical ammunition endpoint damage test, for example, to assess the fragmentation kill range of a warhead, the structural components are uniformly deployed around a circular track with a diameter of approximately 50 meters, and the 0° position of the structural components is designated as the time reference and azimuth reference point.

[0032] Each structural component is quickly snapped into the pre-set ground tripod via a V-shaped quick-release guide rail on its bottom, ensuring stable and precise positioning.

[0033] A target area was set in the center of the site as the core point where the explosion would occur.

[0034] This system employs a highly efficient two-stage rapid extrinsic parameter calibration method to complete the accurate calibration of all structural components within 30 minutes. Internal calibration, performed before the structural components leave the factory, is conducted in a controlled laboratory environment using traditional marking methods such as the Zhang Zhengyou calibration method based on a checkerboard pattern. This process precisely calibrates the two high-speed cameras 2 inside each structural component, obtaining a fixed stereo baseline between them. .

[0035] Simultaneously, the relative pose between each high-speed camera 2 and the built-in inertial measurement unit 3 is calibrated. Fixed three-dimensional baseline and relative pose Once the structural components are manufactured and cured, they are stored in the internal memory of each component. This eliminates the need to repeat this step during field deployment, saving a significant amount of time.

[0036] After the array-level calibration and structural components are deployed, a miniature handheld ball with built-in LEDs is thrown into the center of the field. The ball is gently thrown, allowing it to fall freely in the air and roll randomly for about 5 seconds.

[0037] High-speed cameras 2 in all structural components synchronously acquire motion videos of the LED sphere, ensuring at least 30 valid images of the sphere's center. At the same time, the built-in inertial measurement unit 3 in each structural component also synchronously records its own absolute attitude. In some examples, this absolute attitude is the absolute attitude relative to the direction of gravity and the magnetic north direction.

[0038] The system software receives all camera images and inertial measurement unit 3 data, and optimizes them using a "sphere-IMU" joint adjustment algorithm. This algorithm minimizes the following cost function: ; Among them, the visual reprojection error term : ; in, Let be the pixel coordinates of the center of the sphere of the j-th camera in the k-th frame of the i-th structural component. The system calculates its coordinates with the coordinates of the sphere's center in the k-th frame. π(...) is the camera projection model, and the system calculates... The squared difference of the L2 norm of the pixel coordinates of the projection model π(...) through the camera, where, Let J be the intrinsic parameter matrix of camera j. , World coordinate system to camera The transformation matrix, The pose of the structural component to be optimized. The known quantities are those specified by the manufacturer. Let be the three-dimensional coordinates of the center of the sphere in the k-th frame of the world coordinate system. This is a camera projection model.

[0039] It should be noted that in this embodiment, the projection model π(...) represents the camera projection model, which includes distortion correction and describes how a point in space is mapped to the camera pixel coordinate system plane.

[0040] Specifically, assuming the coordinates of a 3D point in the camera coordinate system are... Then the projection model π ( The calculation process for ) is as follows: Normalized plane projection: ; ; Distortion correction, further, in this embodiment, considering the high precision requirements of high-speed cameras, includes radial distortion and tangential distortion: ; in, , where k is the radial distortion coefficient and p is the tangential distortion coefficient.

[0041] Pixel coordinate mapping: ; ; in, , focal length, , The coordinates of the main point.

[0042] Therefore, in this embodiment, π(...) awards the aforementioned three-dimensional points. Convert to pixel coordinates ( Mathematical functions of .

[0043] Furthermore, in this embodiment, the pose of the structural component to be optimized... This refers to the transformation matrix from the world coordinate system to the IMU coordinate system of the i-th structural component. Specifically, the pose of the structural component to be optimized represents the specific position (translation t) and orientation (rotation R) of each "structural component" in the target range. The specific mathematical expression of the pose of the structural component to be optimized is as follows: ; Wherein, rotation matrix The top left corner of the corresponding matrix Region, describing the three-dimensional orientation of the structural component's coordinate system relative to the world coordinate system; translation vector. The upper right corner of the corresponding matrix Region, i.e. ( This describes the three-dimensional spatial coordinates of the structural component's center in the world coordinate system or the three-dimensional spatial coordinates of the IMU in the world coordinate system; [0,0,0,1] represents the combination of rotation and translation in the same linear transformation matrix, allowing coordinate transformations to be performed through matrix multiplication. , The values ​​t and t are the variables that the optimization algorithm needs to continuously adjust to minimize the error variables.

[0044] The known quantities specified at the factory are the relative extrinsic parameters between the IMU and the camera. Specifically, they describe the fixed position parameters of the camera's optical center relative to the IMU center within the structural components, including a rotation matrix. and Meanwhile, since the camera and IMU are both rigidly mounted in the same "two-piece shock-resistant titanium alloy housing", once calibrated in the factory, this relative relationship will not change regardless of the target range. Therefore, this factory calibration is a known quantity.

[0045] The absolute attitude constraint terms of the inertial measurement unit 3 Defined as: ; Where R_world2imu_i_measured is the absolute rotation matrix directly measured by inertial measurement unit 3, R_world2imu_i_optimized is the rotation matrix part of the optimization variables, and Log(...) is a vector that maps the difference of the rotation matrix to the Lie algebra and calculates its norm.

[0046] It should be noted that the absolute rotation matrix R_world2imu_i_measured and the rotation matrix The rotation directions of the two are the same, the difference lies in the source and purpose of the data. Both are from the world coordinate system to the structural IMU coordinate system, with the starting point being the world 1 coordinate system. In this embodiment, this is a globally fixed reference system determined by the direction of gravity and the magnetic north direction. The structural IMU coordinate system is the coordinate system of the inertial measurement unit fixed inside the device.

[0047] Although the absolute rotation matrix and the rotation matrix describe the same physical transformation direction, in the joint adjustment, the absolute rotation matrix R_world2imu_i_measured represents the observed values, directly calculated from the readings of the IMU's internal accelerometers, gyroscopes, and magnetometers. In the optimization formula, it is a known quantity used to constrain the optimization results from deviating too far from the sensor readings. The rotation matrix is ​​an optimization variable, a value calculated and adjusted during the joint adjustment process. Its purpose is to calculate the difference between the two and try to make the difference as small as possible, but without completely copying the measured value.

[0048] Furthermore, in this embodiment, the optimization variables refer to the variables that the system attempts to adjust to minimize the total error throughout the entire "sphere center-IMU joint adjustment" algorithm. These parameters are the optimization variables. In this embodiment, the optimization variables mainly include the external poses of all structural components and the three-dimensional coordinates of the sphere center, wherein: External orientation of all structural components This refers to the specific position and orientation of each structural component in the world coordinate system.

[0049] The three-dimensional coordinates of the sphere's center refer to the spatial position of the LED sphere used for calibration during the capture of each frame of the image.

[0050] To further clarify, R_world2imu_i_optimized is only the rotation component in the optimization variable of the external pose of the structural component.

[0051] Prior constraints Optionally, one could add, for example, the assumption that the ball's motion is smooth and continuous, or that the structural elements are positioned approximately around the center of the field.

[0052] Furthermore, in some embodiments, Defined as: ; in, and These are weighting coefficients used to balance the importance of different constraint terms in the total cost function.

[0053] in, The constraint on the smoothness of the sphere's motion mainly penalizes and minimizes the non-smooth parts of the trajectory, such as excessive acceleration or acceleration rate. The process of throwing the sphere for approximately 5 seconds generates a series of three-dimensional coordinates of the sphere's center in the world coordinate system. , where k represents the d-th frame of the video.

[0054] In some embodiments, the smoothness constraint formula is the minimum trajectory acceleration or the sum of squares of jerk.

[0055] The formula for minimizing acceleration based on a constant velocity model assumes the sphere tends towards uniform linear motion, and any change in velocity should be penalized, resulting in acceleration. It can be approximated in discrete time as: ; in, It is the time interval between two frames; therefore, the formula for the motion smoothness constraint term can be defined as: ; The formula for minimizing jerk based on a lateral acceleration model assumes the sphere moves in a gravitational field. This model assumes the acceleration changes smoothly, and the jerk... It can be approximated as: ; Constraints are ; More specifically, since the scenario involves free fall and rolling, this embodiment selects a model based on lateral acceleration that minimizes jerk.

[0056] in, This is a weak constraint on the position of structural components, which assumes the position of all i structural components. They are distributed within a roughly circular area with a radius of R centered on the site's center.

[0057] In this embodiment, The formula is defined as the sum of the squares of the differences between the distance of each structural member to the site center and the desired radius R: ; Through optimized solution, the system can quickly calculate the precise external attitude of all structural components in the world coordinate system within minutes. This method can effectively control the 3D reconstruction error of the sphere center and the misalignment of the array stitching.

[0058] Please refer to the instruction manual appendix. Figure 1-5 This invention provides a third embodiment of a scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system. The embodiment provides a method for extracting all fragment parameters and reconstructing three-dimensional kill field, which includes the following steps: S1, High-speed sequence image preprocessing: After the explosion test, the original image sequences acquired by all high-speed cameras 2 are imported into the system. The preprocessing module uses the parallel computing power of the GPU to perform real-time denoising, enhancement, and distortion correction on the images.

[0059] Furthermore, Gaussian filtering can be used for noise reduction, and contrast enhancement can be used for enhancement.

[0060] S2, full-index fragment extraction: S201, Fragment Detection: The preprocessed image is fed into a real-time detection algorithm based on lightweight YOLO-FragNet. This algorithm can quickly and accurately identify fragments in each frame of the image.

[0061] S202, 3D Coordinates and Trajectory: For detected fragments, the system utilizes observation data from multiple high-speed cameras 2 at different perspectives, employing stereo matching and triangulation methods to accurately calculate the 3D spatial coordinates of each fragment. Subsequently, combined with a temporal Kalman filter algorithm, the motion of the fragments between consecutive frames is tracked to construct their complete 3D motion trajectory.

[0062] S203, velocity calculation, for each fragment, calculates the precise three-dimensional velocity vector of the fragment by analyzing its three-dimensional displacement and time interval between consecutive frames, and further improves the accuracy through sub-pixel fitting, effectively controlling the error.

[0063] S204, Size Calculation: For prefabricated fragments with known shape and size, the system directly converts them into actual size by analyzing the contour pixels of the prefabricated fragments in the image.

[0064] For irregularly shaped fragments produced by natural explosions, the system uses a statistical quality model combining the major and minor axes of an equivalent ellipse to estimate the equivalent cubic side length of the fragment, so as to uniformly measure the fragment size.

[0065] S205, density statistics: The system divides a 1m1m1m voxel grid in three-dimensional space. For each voxel, the number of fragments contained therein is counted to ensure that the local fragment density is greater than 25 fragments / ㎡. In areas with sparse fragments, the Kriging interpolation algorithm is used to fill in the data points to ensure the continuity of the kill field.

[0066] S3, 3D Kill Field Reconstruction and Animation Demonstration: S301 imports the extracted fragment point cloud data with attributes such as "position + velocity + size + trajectory" into a 3D engine such as Unity3D or Unreal in real time.

[0067] The system utilizes a GPU particle system to simulate and generate a 0-5 second fragment scattering animation based on the actual velocity vector and trajectory of each fragment. Users can replay the animation from any perspective in the virtual scene, and perform operations such as pausing, rotating, and segmenting to intuitively observe the fragment scattering process.

[0068] The animated demonstration interface can also overlay and display the velocity decay curve of the fragments in real time, and supports users to specify any azimuth / elevation angle slices to analyze the fragment distribution and energy loss in a specific direction.

[0069] Please refer to the instruction manual appendix. Figure 1-5 This invention provides a fourth embodiment of a scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system. The embodiment provides a damage quantification assessment method, which includes the following steps: The S1 features a built-in vulnerability database, and its damage quantification assessment module contains a comprehensive database of vulnerability data for targets such as personnel, vehicles, and radar. This database includes key parameters such as the critical penetration velocity and required mass for various targets.

[0070] S2, Fragment-Target Collision Detection: Users can define or import target models to be evaluated in the 3D kill field animation. The system performs high-precision collision detection based on the 3D trajectory of the fragments and the target model to identify which fragments may hit the target.

[0071] S3, Damage Index Calculation: For each fragment that hits the target, the system combines its velocity, size, and data from the target vulnerability database to calculate the target's "hit count" and "penetration probability".

[0072] For example, if the kinetic energy of a fragment exceeds the critical penetration energy of the target area, then the fragment is considered to be likely to penetrate the target.

[0073] S4. Results Output and Visualization: The system ultimately outputs core damage indicators including kill radius, kill area, and expected damage probability. These results are presented not only in detailed tables but also visualized through charts and 3D heatmaps. The 3D heatmap can intuitively display the degree of damage in different areas and supports bidirectional interaction with the charts. When a user clicks on a data point in the chart, the damage status of the corresponding area is highlighted in the 3D model.

[0074] Specifically, the animation demonstration interface can also directly display specific damage results such as drop velocity, drop angle, explosion height, dispersion angle, envelope area, damage area, and kill radius.

[0075] Furthermore, in order to obtain extremely high-precision sphere center coordinates during array-level calibration, a trajectory extraction method combining Kalman filtering and gray-level weighted mean shift is adopted. This trajectory extraction method is specifically optimized for the luminous sphere that moves freely in the gravitational field in this invention. The specific process is as follows: Step a: For the time-series images acquired by high-speed camera 2, establish the motion state equation of the sphere on the image plane.

[0076] Let the state vector of the sphere at time k be... for: ; in,( ) represents the position of the sphere's center in the image pixel coordinate system. () represents the corresponding pixel velocity component.

[0077] Step b: Considering that the calibration sphere is mainly affected by gravity in the air, neglecting minor air resistance, and given the extremely high frame rate of the high-speed camera, we can approximate the sphere's uniformly accelerated linear motion over a very short time, and establish the state transition equation: ; in: A is the state transition matrix: ; The control input is the projection component of gravitational acceleration onto the image plane. : ; The process noise follows a Gaussian distribution N(0, Q).

[0078] Step c, using the posterior state estimate from the previous time step k-1. Calculate the prior state estimate at time k. : ; Based on the predicted location coordinates ( In the current frame image, a fixed-size rectangular region is selected as the region of interest. This rectangular region can be, for example, 50×50 pixels. This step effectively narrows the search range for subsequent image processing and improves the calculation speed.

[0079] Step d: In the captured region of interest, the center of the spherical spot is precisely located using a gray-scale weighted mean drift algorithm. Since the center of the spot formed by the LED on the imaging sensor has the highest brightness and the edges gradually darken, gray-scale weighting can achieve a finer positioning than sub-pixel level.

[0080] Assuming the pixel coordinates within the region of interest are (i, j) and the grayscale value of this pixel is I(i, j), calculate the zeroth moment of the region of interest. and first moment , : ; ; ; Current grayscale centroid coordinates ( )for: ; The calculated ( Set the center of the new search window, redefine the window, and repeat the above calculation until the center movement distance is less than the preset threshold or the maximum number of iterations is reached. The preset threshold can be, for example, […]. Pixels, the coordinates that eventually converge These are the sub-pixel level observation coordinates of the sphere in the current frame.

[0081] Step e involves using the update equation of the Kalman filter to fuse the predicted and observed values ​​to obtain the optimal estimate.

[0082] Calculate Kalman gain : ; Where H is the observation matrix, which represents only the observation positions: ; R is the observation noise covariance matrix.

[0083] By correcting the prior estimate, we obtain the posterior state estimate. : ; Update error covariance matrix : ; Ultimately, The position component in the image is used as the final sphere center coordinates for the current frame, and... and The data is passed to the next frame (k+1) for recursive calculation. Through this closed-loop correction mechanism, image noise can be effectively suppressed, and high tracking accuracy can still be maintained even when the calibration ball moves quickly or has some motion blur.

[0084] Through the above embodiments, the system of the present invention can efficiently, accurately, and intuitively complete the entire process of ammunition endpoint damage testing, providing powerful data support and evaluation tools for military scientific research and weapon development.

[0085] 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 or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system, characterized in that, include: At least two structural components are arranged around a circle with a predetermined radius around the center of the site for placing unexploded ordnance. The structural components include a housing (1), at least two high-speed cameras (2) located inside the housing (1) and rigidly mounted on a common base plate, and an inertial measurement unit (3) disposed inside the housing (1). A worm gear-worm hinge structure (4) is connected between the optical axes of the high-speed camera (2), which allows the included angle θ between the two optical axes to be continuously adjustable within the range of 15° to 150°; the system also includes three-dimensional dynamic reconstruction and damage assessment software, through which the system performs the following steps: S1, perform internal parameter calibration and external parameter calibration on the structural components on the circumference, and obtain the internal parameters of each high-speed camera (2) and the external attitude of each structural component in the world coordinate system; S2, control the high-speed camera (2) in the structure to synchronously capture the ammunition explosion process at the center of the site and acquire high-speed sequence images; S3. Based on the high-speed sequence images, and combined with the intrinsic parameters and external pose obtained from calibration, solve the motion parameters of each exploded fragment in the visual scene. The motion parameters include at least three-dimensional coordinates, real-time velocity vector, size, and motion trajectory. S4. Based on the motion parameters of the explosive fragments, reconstruct the three-dimensional kill field and calculate the damage index, which includes at least the kill radius and kill area.

2. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 1, characterized in that: Step S1 includes at least array-level calibration, wherein the array-level calibration employs a "sphere-center-IMU" joint adjustment method, which includes the following steps: A ball with a light-emitting mark is thrown at the center of the field. The high-speed camera (2) in each structural component is used to synchronously acquire images of the ball's motion, and the inertial measurement unit (3) is used to synchronously record the absolute attitude of the structural component. Construct a cost function that includes a visual reprojection error term and an inertial measurement unit absolute attitude constraint term; Using the external pose of each structural component and the three-dimensional coordinates of the sphere's center as optimization variables, the external pose of all structural components in the world coordinate system is calculated by minimizing the cost function.

3. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 2, characterized in that: The cost function minimized in the joint adjustment method is defined as: ； in, This is the visual reprojection error term, used to constrain the consistency between the reconstructed sphere center 3D coordinates projected back to the image plane and the observed pixel coordinates. The absolute attitude constraint term for the inertial measurement unit (3) is used to constrain the deviation between the optimized rotation matrix and the measured value of the IMU. These are prior constraints.

4. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 3, characterized in that: The visual reprojection tail error term Defined as: ； in, Let be the pixel coordinates of the center of the sphere of the j-th camera in the k-th frame of the i-th structural component. This is a camera projection model that includes distortion correction. Let be the three-dimensional coordinates of the center of the sphere in the k-th frame of the world coordinate system. Let J be the intrinsic parameter matrix of camera j. Let be the three-dimensional coordinates of the center of the sphere in the k-th frame of the world coordinate system. , World coordinate system to camera The transformation matrix, The pose of the structural component to be optimized. The known quantities are those specified by the manufacturer.

5. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 4, characterized in that: The absolute attitude constraint term of the inertial measurement unit (3) Defined as ； Where R_world2imu_i_measured is the absolute rotation matrix directly measured by the inertial measurement unit (3), R_world2imu_i_optimized is the rotation matrix part in the optimization variables, and Log(...) is the rotation vector that maps the difference of the rotation matrix to the corresponding Lie algebra space. This indicates taking the square of the L2 norm of the rotated vector.

6. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 1, characterized in that: The three-dimensional dynamic reconstruction and damage software includes a full-index fragment extraction module. The full-index fragment extraction module uses a lightweight YOLO-FragNet algorithm to detect fragments in real time from preprocessed single-frame images. It uses stereo matching and triangulation to calculate the three-dimensional coordinates of the detected fragments, and combines temporal Kalman filtering to correlate and track the motion of fragments between consecutive frames and obtain the three-dimensional motion trajectory of the fragments.

7. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 6, characterized in that: The 3D dynamic reconstruction and destruction software also includes a 3D kill field reconstruction module. The 3D kill field reconstruction module imports the extracted fragment point cloud data with position, velocity, size and trajectory attributes into the 3D rendering engine. Using a GPU particle system, each fragment particle is driven according to the calculated real velocity vector to generate an interactive 3D animation of fragment scattering.

8. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 1, characterized in that: The three-dimensional dynamic reconstruction and damage assessment software also includes a damage quantification assessment module. The damage quantification assessment module has a built-in vulnerability database containing personnel, vehicles and radar targets. The vulnerability database stores the target's critical penetration velocity and necessary damage mass parameters. The damage quantification assessment module is configured to compare the kinetic energy of the fragments with the parameters in the vulnerability database to calculate the number of hits and the penetration probability.

9. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 2, characterized in that: In the array-level calibration, the method for calculating the two-dimensional trajectory of the sphere in the image using the high-speed camera (2) in each structural component includes the following steps: Step a: For the time-series images acquired by the high-speed camera, establish the motion state vector of the sphere. The motion state vector includes at least the two-dimensional position coordinates and two-dimensional velocity components of the sphere in the image coordinate system. Step b: Using the prediction equation of the Kalman filter, based on the optimal estimated state of the previous moment, and combined with the projection component of the gravitational acceleration on the image plane as the control output, the prior state estimate of the sphere in the current frame is calculated. Step c: Based on the position coordinates in the prior state estimate, extract a rectangular region centered at that position as the region of interest in the current frame image.

10. The scalable surround dual-camera array fragment kill field three-dimensional reconstruction damage assessment system according to claim 9, characterized in that: The method for calculating the two-dimensional trajectory of a sphere in an image further includes the following steps: Step d: In the region of interest, the gray-scale weighted mean drift algorithm is used to locate the center of the spherical spot. The gray value of each pixel in the region of interest is used as the weight to calculate the zero-order moment and the first-order moment of the region. The gray-scale centroid is obtained by dividing the first-order moment by the zero-order moment. The search is iteratively searched with this gray-scale centroid as the center until convergence, and the sub-pixel level observation target of the sphere is obtained. Step e: Using the sub-pixel level observation coordinates as measurement vectors, the prior state estimate is corrected using the update equation of the Kalman filter to obtain the posterior state estimate of the current frame, and the posterior state estimate is used as the input for prediction of the next frame, until the trajectory extraction of the entire sequence image is completed.

Images