A method and system for detecting four-dimensional motion trajectories of a group of fragments under an event camera

Abstract

This invention discloses a method and system for detecting the four-dimensional motion trajectory of a fragment group under an event camera. The method includes: acquiring an asynchronous event stream of a fragment explosion scene using an event camera; performing polarity-guided adaptive spatiotemporal trajectory manifold fitting, constructing a spatiotemporal clustering and regression model based on event polarity features, mapping the discrete asynchronous event stream into the temporal projection trajectory of fragments on the image plane; and performing multi-view spatiotemporal manifold collaborative fragment four-dimensional trajectory reconstruction. Based on the temporal projection trajectory under multiple views, through cross-view trajectory association and reprojection optimization, the three-dimensional spatial position and motion parameters of the fragments evolving over time are calculated, and the camera pose is dynamically compensated simultaneously, ultimately outputting the four-dimensional motion trajectory of the fragment group. This invention constructs a multi-view spatiotemporal manifold collaborative four-dimensional trajectory reconstruction model, simultaneously realizing dynamic compensation for camera extrinsic parameter deviations and refinement of motion parameters, thereby calculating a high-precision four-dimensional motion trajectory of the fragment group.

Description

Technical Field

[0001] This invention relates to the field of camera motion trajectory detection technology, and in particular to a method and system for detecting the four-dimensional motion trajectory of fragment groups under an event camera. Background Technology

[0002] Fragmentation warheads, as a type of weapon configuration that achieves mass destruction by driving a large number of high-speed discrete fragments through explosion, depend on the spatiotemporal distribution characteristics of the fragmentation field for their combat effectiveness. The initial velocity distribution, dispersion angle, and shape characteristics at the moment of explosion directly determine the warhead's kinetic energy output, penetration capability, and hit probability. Therefore, obtaining accurate and time-series continuous fragment kinematic data not only constitutes the basic input for combat effectiveness assessment, structural optimization, pre-formed fragment arrangement, and the establishment of offensive and defensive models, but also serves as a key basis for test result reproduction, product type approval and certification, and accident forensic analysis. In conclusion, fragment kinematic parameter measurement technology has irreplaceable strategic significance and engineering value in defense research and development, testing and verification, and equipment life-cycle assessment.

[0003] Existing fragment measurement methods are mainly divided into two categories: contact and non-contact. Contact methods directly acquire fragment information through capture nets, impact plates, or force sensors. These methods are relatively simple to operate and provide intuitive energy measurements, but they are prone to disturbing fragment motion. In high-density, high-speed scenarios, sample loss or measurement saturation can occur, and spatial coverage and data repeatability are limited. In contrast, non-contact methods, such as Doppler radar, flash X-rays, and light curtain targets, overcome the limitations of traditional contact methods in terms of measurement range and recording continuity without interfering with fragment motion. They utilize measurement mechanisms such as photoelectric sensing, electromagnetic echo, or radiation imaging to achieve accurate measurement of high-dimensional features such as fragment velocity vectors, spatial distribution density, and flight attitude. Compared to other non-contact methods, visual measurement has significant advantages such as simple deployment, wide field of view coverage, and strong visualization capabilities, making it one of the most popular non-contact techniques in modern fragment kinematics research.

[0004] However, traditional high-speed cameras, as the primary carrier of visual measurement, are still prone to problems such as image saturation, motion blur, and heavy data burden in high-speed moving target measurement scenarios, thus limiting their application in high-speed fragment measurement to some extent. In recent years, event cameras based on neuromorphic vision principles have attracted widespread attention as a novel visual sensor. They record brightness change events asynchronously at the pixel level, featuring microsecond-level temporal resolution, high dynamic range, and low data redundancy. They can effectively acquire instantaneous target motion information under extreme lighting and high-speed motion conditions, providing a new technical approach for high-speed fragment trajectory measurement and motion parameter calculation. However, event stream data has characteristics such as asynchronous triggering, sparse spatial distribution, and unstructured representation, making it difficult to directly form stable and continuous target representations. In dense fragment explosion scenarios, due to the large number of targets, high speed, and frequent trajectory intersections, the events corresponding to multiple targets highly overlap in the spatiotemporal domain, leading to difficulties in separating single-target events and unclear cross-view correspondences. At the same time, affected by environmental disturbances and event noise, the reconstruction accuracy further decreases during parameter optimization, making high-precision positioning and continuous trajectory reconstruction of fragment groups still face significant challenges. Therefore, a method and system for detecting the four-dimensional motion trajectory of fragment groups under an event camera is needed. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a method and system for detecting the four-dimensional motion trajectory of a fragment group under an event camera.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention includes the following steps: S1: Acquire an asynchronous event stream of the fragment explosion scene through an event camera. The asynchronous event stream includes pixel coordinates, polarity of light intensity change, and microsecond-level timestamps. S2: Perform polarity-guided adaptive spatiotemporal trajectory manifold fitting, construct spatiotemporal clustering and regression models based on event polarity features, and map discrete asynchronous event streams into temporal projection trajectories of fragments on the image plane; S3: Perform multi-view spatiotemporal manifold collaborative fragment four-dimensional trajectory reconstruction. Based on the temporal projection trajectory under multiple views, calculate the three-dimensional spatial position and motion parameters of the fragments as they evolve over time through cross-view trajectory association and reprojection optimization, and simultaneously perform dynamic compensation on the camera pose, finally outputting the four-dimensional motion trajectory of the fragment group.

[0008] Furthermore, the polarity-guided adaptive spatiotemporal trajectory manifold fitting includes: S21: Spatiotemporal manifold clustering with coupled polarity features. Spatiotemporal density descriptors are constructed based on polarity-induced spatiotemporal distance. Structured target clusters are obtained through core manifold detection and spatiotemporal connectivity expansion, and isolated noise events are filtered out. S22: Bipolar coherence-guided RANSAC trajectory fitting, constructing a dual-track collaborative trajectory regression model based on the spatiotemporal centroid of the target cluster to generate candidate projection trajectories for fragments; S23: Multi-criteria consistency trajectory deep refinement, through merging of homogeneous trajectories, kinematic feasibility filtering, and geometric smoothness constraints, to obtain the refined fragment projection trajectory sequence.

[0009] Furthermore, the polarity-induced spatiotemporal distance is used to balance the order-of-magnitude difference between spatial pixel coordinates and microsecond-level timestamps; the asynchronous event stream is deconstructed into independent temporal subspaces through temporal domain segmentation, and then a local spatiotemporal manifold density descriptor is constructed.

[0010] Furthermore, in step S22, the constructed dual-track collaborative regression model is as follows:

[0011] Where A, B, and C are parameter vectors characterizing the common motion characteristics of the fragments. To describe the spatial displacement vector of polarity edge bias, a joint evaluation index for candidate trajectories is established through a noise scale edgeification mechanism, and an adaptive discrimination threshold that maximizes the inter-class variance is used to screen out effective candidate projected trajectories from the candidate trajectory hypotheses.

[0012] Furthermore, the fragment four-dimensional trajectory reconstruction based on multi-view spatiotemporal manifold collaboration includes: S31: Cross-view trajectory association reconstruction based on epipolar voting mechanism, extracting discrete feature point pairs from multi-view refined projection trajectory, completing cross-view trajectory matching through epipolar voting, and solving the three-dimensional coordinate sequence based on spatial intersection. S32: Four-dimensional trajectory regression with coupled acceleration features. Construct a four-dimensional trajectory regression model including initial velocity, acceleration, and initial position to solve for the initial motion parameters of the fragment. S33: Global trajectory refinement that couples polarity bias with pose perturbation, constructs an event-level reprojection cost function, and iteratively optimizes and compensates for camera pose deviations and refines trajectory parameters.

[0013] Furthermore, multiple sets of discrete feature point pairs are extracted at equal intervals within the co-temporal domain. Based on the fundamental matrix of the binocular system and the epipolar search bandwidth, the number of valid votes for observation points on candidate trajectories falling within the corresponding epipolar search band is counted as the trajectory similarity score. The trajectory pair with the highest score from different perspectives is selected as the matching result. Furthermore, the polarity offset coefficient and the camera pose perturbation matrix are incorporated into a unified optimization framework. The damped least squares algorithm is used to solve for the optimal motion parameters and pose perturbation parameters, and the inlier set is dynamically updated until convergence.

[0014] Further, in step S22: two support points are randomly selected from the spatiotemporal centroid sequences corresponding to ON polarity and OFF polarity to construct a cross-polarity minimum sampling set, and the motion vector and spatial displacement vector are solved simultaneously to generate candidate trajectory hypotheses.

[0015] Furthermore, in step S33: the event-level reprojection cost function takes the sum of squared residuals between the actual observed coordinates of the event and the corresponding theoretical projection points as the optimization objective; after each round of parameter update, the set of interior points is dynamically resampled through the reprojection distance threshold criterion, and the cost function is iteratively driven to converge.

[0016] On the other hand, a four-dimensional motion trajectory detection system for a fragment group under an event camera, used to execute the aforementioned four-dimensional motion trajectory detection method for a fragment group under an event camera, includes: The event acquisition module is used to acquire asynchronous event streams of fragment explosion scenes through an event camera. The asynchronous event streams include pixel coordinates, polarity of light intensity changes, and microsecond-level timestamps. The polarity-guided trajectory fitting module is used to perform polarity-guided adaptive spatiotemporal trajectory manifold fitting, which transforms discrete asynchronous event streams into temporal projection trajectories of fragments on the image plane. The four-dimensional view reconstruction module is used to perform fragment four-dimensional trajectory reconstruction by multi-view spatiotemporal manifold collaboration. Based on the temporal projection trajectory under multiple views, it calculates the three-dimensional spatial position and motion parameters of the fragments as they evolve over time through cross-view trajectory association and reprojection optimization, and simultaneously performs dynamic compensation for camera pose, finally outputting the four-dimensional motion trajectory of the fragment group.

[0017] The beneficial effects of this invention are: This invention addresses the problem of unstructured feature extraction from asynchronous event streams by employing polarity-guided adaptive spatiotemporal manifold fitting technology, mapping discrete event sequences into deterministic projected motion trajectories with physical evolution laws. Furthermore, it utilizes a multi-view spatiotemporal manifold collaborative fragment four-dimensional trajectory reconstruction technology. By minimizing the reprojection cost function of coupled trajectory parameters and pose perturbations, it simultaneously achieves target four-dimensional trajectory reconstruction and dynamic refinement of camera extrinsic parameters. This significantly improves the accuracy of motion parameters and four-dimensional trajectory measurement for high-speed targets in long-term dynamic environments and effectively suppresses measurement accumulation errors caused by system geometric instability. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the principle of the four-dimensional motion trajectory detection method for fragment groups based on an event camera array as described in this invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] like Figure 1 As shown, the present invention includes the following steps: S1: Acquire an asynchronous event stream of the fragment explosion scene through an event camera. The asynchronous event stream includes pixel coordinates, polarity of light intensity change, and microsecond-level timestamps. S2: Perform polarity-guided adaptive spatiotemporal trajectory manifold fitting, construct spatiotemporal clustering and regression models based on event polarity features, and map discrete asynchronous event streams into temporal projection trajectories of fragments on the image plane; S3: Perform multi-view spatiotemporal manifold collaborative fragment four-dimensional trajectory reconstruction. Based on the temporal projection trajectory under multiple views, calculate the three-dimensional spatial position and motion parameters of the fragments as they evolve over time through cross-view trajectory association and reprojection optimization, and simultaneously perform dynamic compensation on the camera pose, finally outputting the four-dimensional motion trajectory of the fragment group.

[0023] The polarity-guided adaptive spatiotemporal trajectory manifold fitting includes: S21: Spatiotemporal manifold clustering with coupled polarity features. Spatiotemporal density descriptors are constructed based on polarity-induced spatiotemporal distance. Structured target clusters are obtained through core manifold detection and spatiotemporal connectivity expansion, and isolated noise events are filtered out. S22: Bipolar coherence-guided RANSAC trajectory fitting, constructing a dual-track collaborative trajectory regression model based on the spatiotemporal centroid of the target cluster to generate candidate projection trajectories for fragments; S23: Multi-criteria consistency trajectory deep refinement, through merging of homogeneous trajectories, kinematic feasibility filtering, and geometric smoothness constraints, to obtain the refined fragment projection trajectory sequence.

[0024] The polarity-induced spatiotemporal distance is used to balance the order-of-magnitude difference between spatial pixel coordinates and microsecond-level timestamps; the asynchronous event stream is deconstructed into independent temporal subspaces through temporal domain segmentation, and then a local spatiotemporal manifold density descriptor is constructed.

[0025] The camera pose perturbation matrix is ​​used to dynamically correct the camera reference extrinsic matrix, thereby compensating for the temporal drift of the camera extrinsic caused by the explosion impact; and the polarity bias correction mapping is used to eliminate the geometric center offset of the fragments caused by the edge imaging characteristics of the event camera.

[0026] The homogeneous trajectory merging is used to calculate the correlation coefficient of motion parameters between different candidate trajectories and merge trajectory segments representing the same physical target to eliminate duplicate detection; the kinematic feasibility filtering is used to solve the average velocity according to the trajectory equation and eliminate abnormal motion trajectories based on the preset velocity domain constraint; the geometric smoothness constraint is used to evaluate the trajectory shape using the trajectory curvature criterion, eliminate trajectory nonlinear distortion and ensure that the trajectory conforms to the ballistic smooth evolution law.

[0027] (1) Polarity-guided adaptive spatiotemporal trajectory manifold fitting When using an event camera to acquire information about the dynamic changes of fragments within the field of view, the output data is an asynchronous event stream sequence. Given that the original event stream exhibits discrete and unstructured pixel-level responses, it cannot directly represent the overall geometry and motion patterns of the fragments and is susceptible to background noise. Therefore, it is necessary to aggregate spatiotemporal features of valid events and further abstract the aggregated discrete target state into a continuous parametric model. Thus, a fragment projection trajectory fitting strategy based on spatiotemporal clustering and linear regression needs to be constructed to achieve accurate mapping from the asynchronous event stream to a deterministic motion trajectory.

[0028] First, define It is an asynchronous event stream captured by the event camera, in which, Represents the pixel coordinates of the event. Polarity characteristics (ON / OFF) indicating changes in light intensity. It is a microsecond-level timestamp.

[0029] ① Spatiotemporal manifold clustering based on coupled polarity features: Two events are defined based on the consistent physical properties of event polarity generated by fragments during high-speed motion. and Polarity-induced spatiotemporal distance between :

[0030] in, This is the spatiotemporal mapping weight coefficient, used to balance the order-of-magnitude difference between spatial pixel coordinates and microsecond-level timestamps.

[0031] To suppress clustering and adhesion caused by the intersection of multiple fragment trajectories, time-domain segmentation is used to deconstruct the asynchronous event stream into independent time-domain subspaces, and within each time window, the event stream is further segmented based on spatiotemporal distance. Constructing local spatiotemporal manifold density descriptors :

[0032] Based on the constructed spatiotemporal density field characterization, the mapping of discrete asynchronous event points to structured target clusters is achieved through core manifold detection and spatiotemporal manifold connectivity expansion.

[0033] Core manifold detection: Traverse the event sequence, if Meets minimum density threshold Then The event is identified as a core event with motion semantics, and a spatiotemporal cluster is initiated using this event as a seed point. The evolution of.

[0034] Spatiotemporal manifold expansion and noise reduction: Using density reachability logic, all event points in the neighborhood that meet the polarity consistency are merged into the manifold. During the iteration process, isolated events whose spatiotemporal density cannot be reached through any core point are identified as background thermal noise or random clutter and filtered out.

[0035] ② Bipolar coherent guided RANSAC trajectory fitting: The clustered RANSAC trajectories are fitted with the RANSAC trajectories. polar clusters and Polar clusters are mapped to a unified time axis, and the spatiotemporal centroid sequence of each polar cluster is extracted. Utilizing the parallel evolution of fragment edges within the spatiotemporal manifold, a shared dynamic parameter and a fixed spatial offset are introduced to establish the following dual-track coordinated trajectory regression model:

[0036] Where A, B, and C are parameter vectors characterizing the common motion characteristics of the fragments. This describes the spatial displacement vector of the polarity edge offset.

[0037] For the extracted asynchronous spatiotemporal centroid sequence and Two support points are randomly selected from each set to construct a transpolar minimum sampling set, and the motion vector is solved synchronously. With spatial displacement vector Candidate spatiotemporal trajectory hypotheses are obtained, and a noise scale marginalization mechanism is introduced to establish a joint evaluation index for candidate trajectories. :

[0038] in, The noise scaling variable represents the deviation of the centroid from the tolerance range of the ideal trajectory. For this noise scale in continuous scale space The prior probability distribution within.

[0039] Scoring sets of trajectory hypotheses generated on a large scale Traverse the score space to obtain the inter-class variance Maximize the adaptive discrimination threshold :

[0040] in, , Scores below and above the thresholds, respectively. The probability weights of the candidate model set. , These are the average scores for the corresponding model sets.

[0041] If the evaluation index of the candidate trajectory hypothesis Exceeding this adaptive threshold If so, it is confirmed as a valid candidate model. Furthermore, by eliminating polar space bias... By regressing the observations from the dual-track edge to the physical centroid, candidate projected trajectory models representing the true motion of the fragment are obtained:

[0042] ③ Deep refinement of multi-criteria consistency trajectories: For the acquired candidate projection trajectory models, a multi-criteria consistency verification mechanism based on topology, dynamics, and geometry is introduced to achieve in-depth refinement of the models, and finally outputs a projection trajectory sequence with spatiotemporal continuity and high reliability.

[0043] Homogeneous trajectory merging: Primarily used to eliminate duplicate detections. Utilizes motion parameter vectors. Calculate the correlation coefficient between candidate trajectories The parameters of the same source segments are integrated to obtain a global continuous motion description of the target.

[0044]

[0045] Kinematic feasibility filtering: Primarily used to eliminate flicker noise and linear artifacts. It calculates the average velocity of the trajectory within the field of view by solving the trajectory equation. Furthermore, velocity domain constraints are set based on prior dynamics to eliminate abnormal motion descriptions.

[0046]

[0047] in, and These represent the start and end observation times of the trajectory segment in the image coordinate system, respectively.

[0048] Geometric smoothness constraints: Primarily used to eliminate trajectory nonlinear distortions caused by local impulse noise. Utilizing curvature criteria. Evaluate the instantaneous geometry of the trajectory to ensure that the output trajectory conforms to the smooth evolution of the ballistic trajectory.

[0049]

[0050] (2) Fragment four-dimensional trajectory reconstruction by multi-view spatiotemporal manifold collaboration To address the challenges of highly overlapping targets in dense explosion fields, the unstructured nature of edge imaging by event cameras, and the temporal drift of camera pose caused by explosion impact, this solution constructs a fragment four-dimensional trajectory reconstruction method based on spatiotemporal manifold collaboration. While achieving high-precision calculation of motion parameters such as initial velocity, acceleration, and initial position of fragments, it simultaneously completes dynamic compensation for camera array pose deviation, thereby obtaining a high-precision four-dimensional motion trajectory sequence with global spatiotemporal consistency in extreme dynamic environments.

[0051] ① Cross-view trajectory association reconstruction based on epipolar zone voting mechanism: For the left and right view trajectories after spatiotemporal alignment, 50 sets of discrete feature point pairs are extracted at equal intervals within the co-temporal domain to construct a cross-view spatiotemporal observation sequence. An epipolar band-constrained voting mechanism is introduced to count the number of valid votes for corresponding points in the candidate trajectories of the right view that fall within the epipolar search band, thereby obtaining a similarity score that measures the strength of the association between trajectories from different viewpoints. :

[0052] Where F is the fundamental matrix of the binocular system. The bandwidth for epipolar search is defined as follows. Finally, the trajectory pair with the highest similarity score from different viewpoints is selected as the matching result, and the three-dimensional coordinate sequence of the target at each sampling time is calculated using the principle of spatial intersection. .

[0053] To further obtain a description of the spatiotemporal continuity of the target, a four-dimensional trajectory regression model with coupled acceleration features is established:

[0054] in, It is the acceleration vector. The initial velocity vector, Let be the starting position coordinates. The initial motion parameter vector is solved by minimizing the sum of squared residuals of the distances from each sampling point to the parameterized curve at each time step:

[0055] Ultimately, the discrete point cloud observed from different perspectives is transformed into the initial spatiotemporal motion equations of the fragments, which possess physical evolution laws.

[0056] ② Global trajectory refinement of coupled polarity bias and pose perturbation: Define the left and right cameras in the binocular system ( The baseline extrinsic parameter matrix is Its dynamic correction process is expressed as ,in To represent the pose perturbation matrix, the initial motion equations are used. And its three-dimensional motion vector, calculate the projected coordinates of the target in real time in each view. With projected velocity vector :

[0057] Within each view, construct a polarity offset vector based on the projection velocity direction to obtain the polarity-corrected mapped projection coordinates:

[0058] in, and Corresponding to polarity and Geometric projection predictions of polarity edge observations. This represents the polarity offset coefficient. To address the system's geometric instability under dynamic environments, trajectory parameters and camera pose perturbation terms are incorporated into a unified optimization framework. An event-level reprojection cost function is constructed, and the optimal motion parameters are obtained using the Levenberg-Marquardt algorithm. and pose perturbation parameters :

[0059] In this reprojection cost function, These are the actual observed coordinates. To determine the polarity of the event The corresponding theoretical projection points are selected. After each round of parameter updates, the system dynamically resamples the set of interior points using the reprojection distance threshold criterion, and iteratively drives the optimization process of the cost function until the solution parameters reach a convergent state.

[0060] Through the above nonlinear iterative solution process, the accurate initial velocity, acceleration and initial position are obtained, and the dynamic compensation of camera pose deviation and the refinement of trajectory description parameters are realized simultaneously. Finally, a high-precision four-dimensional motion trajectory of the fragment group with global spatiotemporal consistency is output.

[0061] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting the four-dimensional motion trajectory of a fragment group under an event camera, characterized in that, Includes the following steps: S1: Acquire an asynchronous event stream of the fragment explosion scene through an event camera. The asynchronous event stream includes pixel coordinates, polarity of light intensity change, and microsecond-level timestamps. S2: Perform polarity-guided adaptive spatiotemporal trajectory manifold fitting, construct spatiotemporal clustering and regression models based on event polarity features, and map discrete asynchronous event streams into temporal projection trajectories of fragments on the image plane; S3: Perform multi-view spatiotemporal manifold collaborative fragment four-dimensional trajectory reconstruction. Based on the temporal projection trajectory under multiple views, calculate the three-dimensional spatial position and motion parameters of the fragments as they evolve over time through cross-view trajectory association and reprojection optimization, and simultaneously perform dynamic compensation on the camera pose, finally outputting the four-dimensional motion trajectory of the fragment group.

2. The method according to claim 1, characterized in that, The polarity-guided adaptive spatiotemporal trajectory manifold fitting includes: S21: Spatiotemporal manifold clustering with coupled polarity features. Spatiotemporal density descriptors are constructed based on polarity-induced spatiotemporal distance. Structured target clusters are obtained through core manifold detection and spatiotemporal connectivity expansion, and isolated noise events are filtered out. S22: Bipolar coherence-guided RANSAC trajectory fitting, constructing a dual-track collaborative trajectory regression model based on the spatiotemporal centroid of the target cluster to generate candidate projection trajectories for fragments; S23: Multi-criteria consistency trajectory deep refinement, through merging of homogeneous trajectories, kinematic feasibility filtering, and geometric smoothness constraints, to obtain the refined fragment projection trajectory sequence.

3. The method for detecting the four-dimensional motion trajectory of a fragment group under an event camera according to claim 2, characterized in that, The polarity-induced spatiotemporal distance is used to balance the order-of-magnitude difference between spatial pixel coordinates and microsecond-level timestamps; the asynchronous event stream is deconstructed into independent temporal subspaces through temporal domain segmentation, and then a local spatiotemporal manifold density descriptor is constructed.

4. The method for detecting the four-dimensional motion trajectory of a fragment group under an event camera according to claim 2, characterized in that, In step S22, the constructed dual-track collaborative regression model is as follows: ； Where A, B, and C are parameter vectors characterizing the common motion characteristics of the fragments. This describes the spatial displacement vector of the polarity edge offset; A joint evaluation index for candidate trajectories is established through a noise scale marginalization mechanism, and an adaptive discrimination threshold that maximizes the inter-class variance is used to screen out effective candidate projected trajectories from the candidate trajectory hypotheses.

5. The method for detecting the four-dimensional motion trajectory of a fragment group under an event camera according to claim 1, characterized in that, The fragment four-dimensional trajectory reconstruction by multi-view spatiotemporal manifold collaboration includes: S31: Cross-view trajectory association reconstruction based on epipolar voting mechanism, extracting discrete feature point pairs from multi-view refined projection trajectory, completing cross-view trajectory matching through epipolar voting, and solving the three-dimensional coordinate sequence based on spatial intersection. S32: Four-dimensional trajectory regression with coupled acceleration features. Construct a four-dimensional trajectory regression model including initial velocity, acceleration, and initial position to solve for the initial motion parameters of the fragment. S33: Global trajectory refinement that couples polarity bias with pose perturbation, constructs an event-level reprojection cost function, and iteratively optimizes and compensates for camera pose deviations and refines trajectory parameters.

6. The method for detecting the four-dimensional motion trajectory of a fragment group under an event camera according to claim 5, characterized in that, Multiple discrete feature point pairs are extracted at equal intervals within the co-temporal domain. Based on the binocular system's fundamental matrix and epipolar search bandwidth, the number of valid votes for observation points on candidate trajectories falling within the corresponding epipolar search band is counted as the trajectory similarity score. The trajectory pair with the highest score from different perspectives is selected as the matching result.

7. The method for detecting the four-dimensional motion trajectory of a fragment group under an event camera according to claim 5, characterized in that, The polarity offset coefficient and the camera pose perturbation matrix are incorporated into a unified optimization framework. The optimal motion parameters and pose perturbation parameters are solved by the damped least squares algorithm, and the inlier set is dynamically updated until convergence.

8. The method for detecting the four-dimensional motion trajectory of a fragment group under an event camera according to claim 2, characterized in that, In step S22: two support points are randomly selected from the spatiotemporal centroid sequences corresponding to ON polarity and OFF polarity to construct a cross-polarity minimum sampling set, and the motion vector and spatial displacement vector are solved simultaneously to generate candidate trajectory hypotheses.

9. The method for detecting the four-dimensional motion trajectory of a fragment group under an event camera according to claim 5, characterized in that, In step S33: the event-level reprojection cost function takes the sum of squared residuals between the actual observed coordinates of the event and the corresponding theoretical projection points as the optimization objective; after each round of parameter update, the set of interior points is dynamically resampled through the reprojection distance threshold criterion, and the cost function is iteratively driven to converge.

10. A four-dimensional motion trajectory detection system for a fragment group under an event camera, used to execute the four-dimensional motion trajectory detection method for a fragment group under an event camera as described in any one of claims 1-9, characterized in that, include: The event acquisition module is used to acquire asynchronous event streams of fragment explosion scenes through an event camera. The asynchronous event streams include pixel coordinates, polarity of light intensity changes, and microsecond-level timestamps. The polarity-guided trajectory fitting module is used to perform polarity-guided adaptive spatiotemporal trajectory manifold fitting, which transforms discrete asynchronous event streams into temporal projection trajectories of fragments on the image plane. The four-dimensional view reconstruction module is used to perform fragment four-dimensional trajectory reconstruction by multi-view spatiotemporal manifold collaboration. Based on the temporal projection trajectory under multiple views, it calculates the three-dimensional spatial position and motion parameters of the fragments as they evolve over time through cross-view trajectory association and reprojection optimization, and simultaneously performs dynamic compensation for camera pose, finally outputting the four-dimensional motion trajectory of the fragment group.

Images