A point cloud registration method and system for power battery reconstruction

Abstract

This invention proposes a point cloud registration method and system for power battery reconstruction, relating to the field of 3D point cloud data processing technology. The method includes: establishing an initial point correspondence between a source point cloud and a target point cloud; using the PSO (Particle Swarm Optimization) algorithm to simulate and update the initial point correspondence, obtaining a coarsely registered first particle swarm; combining the ICP (Integrated Point Optimization) algorithm to minimize the Euclidean distance between corresponding points in the source and target point clouds, determining the optimal rigid transformation, and obtaining a finely registered second particle swarm; during each coarse and fine registration process, if the second fitness value corresponding to the second particle swarm is less than the first fitness value corresponding to the first particle swarm, the second pose of the second particle swarm is fused with the first pose of the first particle swarm to correct the first pose, and the target pose that minimizes the fitness function value is selected from the second pose to optimize the particle poses of the idle particle swarm of the first particle swarm, obtaining the target particle swarm.

Description

Technical Field

[0001] This invention relates to the field of 3D point cloud data processing technology, and in particular to a point cloud registration method and system for power battery reconstruction. Background Technology

[0002] Power batteries are a core component of new energy vehicles, and their quality has a crucial impact on the vehicle's range and safety. To conduct comprehensive quality inspection, fault analysis, and optimized design of power batteries, a complete and accurate 3D model is required. However, due to the complex structure of power batteries and issues such as occlusion, a single scan often only acquires localized 3D information. Therefore, multiple scans are necessary, and point cloud registration technology is used to merge the localized point cloud data obtained from these scans into a complete 3D point cloud, thereby achieving 3D reconstruction of the power battery. Point cloud registration is a critical step in 3D reconstruction, and its accuracy and robustness directly affect the quality of subsequent inspections. Currently, mainstream methods generally employ a two-stage sequential framework of coarse registration followed by fine registration.

[0003] However, the performance of fine registration is highly dependent on the quality of the initial pose provided by coarse registration. When the coarse registration result is unreliable, the fine registration algorithm is prone to getting trapped in local optima, leading to overall registration failure. For example, feature-matching-based coarse registration methods may have reduced matching reliability in scenarios with sparse features or repetitive structures. Classical fine registration algorithms, such as ICP (Iterative Closest Point), are sensitive to the initial pose and are prone to converging to local optima when the initial deviation is large. While global search-based methods, such as PSO (Particle Swarm Optimization), can avoid local optima, they often lack fine-tuning capabilities in the later stages and are difficult to achieve extremely high accuracy independently. The currently widely used "coarse-to-fine" sequential approach often executes the coarse and fine registration stages independently with unidirectional information transmission. If the coarse registration result has a certain deviation, the fine registration stage cannot effectively correct it, thus affecting the robustness and accuracy of the overall registration. Summary of the Invention

[0004] In view of this, the present invention proposes a point cloud registration method and system for power battery reconstruction.

[0005] The technical solution of this invention is implemented as follows: The first aspect of this invention provides a point cloud registration method for power battery reconstruction, comprising: The initial point correspondence between the source point cloud and the target point cloud is established by utilizing the geometric relationship between the feature descriptors constructed by pasting marker points on the surface of the battery to be registered; The PSO (Particle Swarm Optimization) algorithm is used to simulate the correspondence of the initial points. The particle swarm is updated based on the individual optimal position and the global optimal position to obtain the first particle swarm after coarse registration. The ICP (Integrated Point Plane) algorithm is combined to minimize the Euclidean distance between the corresponding points in the source point cloud and the target point cloud to determine the optimal rigid transformation and obtain the second particle swarm after fine registration. In each coarse and fine registration process, the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm are obtained. If the second fitness value is less than the first fitness value, the second pose of the second particle swarm is fused with the first pose of the first particle swarm to complete the correction of the first pose. The target pose that minimizes the fitness function value is selected from the second pose as the global structure guide term to optimize the particle poses of the idle particle swarm of the first particle swarm to obtain the target particle swarm. The idle particle swarm includes particles that did not participate in fine registration.

[0006] Based on the above technical solutions, preferably, the step of establishing the initial point correspondence between the source point cloud and the target point cloud by utilizing the geometric relationship between the feature descriptors constructed using marker points pasted on the surface of the battery to be registered includes: The marked points in the source point cloud and the target point cloud are described by features respectively, and the Euclidean distance between each feature vector is obtained. Using each marker point in the source point cloud as a reference, search for the marker point with the smallest Euclidean distance in the target point cloud to obtain a first matching set; and using each marker point in the target point cloud as a reference, search for the marker point with the smallest Euclidean distance in the source point cloud to obtain a second matching set. The initial point correspondence between the source point cloud and the target point cloud is established based on the intersection of the point clouds of the first matching set and the second matching set.

[0007] Based on the above technical solutions, preferably, the step of establishing the initial point correspondence between the source point cloud and the target point cloud based on the intersection of the first matching set and the second matching set includes: The intersection of the point cloud is solved using the singular value decomposition method to obtain the optimal rigid body transformation parameters; The optimal rigid body transformation parameters are used as initial elite particles, and multiple subsequent elite particles are generated by combining them with uniformly sampled random perturbation vectors. The initial point correspondence between the source point cloud and the target point cloud is established based on the initial elite particles and the subsequent elite particles.

[0008] Based on the above technical solutions, preferably, the step of simulating the initial point correspondence using the PSO particle swarm optimization algorithm, and updating the initial point correspondence based on the individual optimal position and the global optimal position of the particle swarm to obtain a first particle swarm after coarse registration includes: The current position of a particle is updated based on its historical velocity, historical position, historical individual best position, and historical group best position, to obtain the position with the minimum fitness value in this round. If the fitness value corresponding to the current position is less than the fitness value of the historical individual's best position, the current position replaces the historical individual's best position; otherwise, the historical individual's best position remains unchanged.

[0009] Based on the above technical solutions, preferably, before minimizing the Euclidean distance between corresponding points in the source point cloud and the target point cloud using the ICP algorithm, determining the optimal rigid transformation, and obtaining the finely registered second particle swarm, the following steps are also included: Obtain the residual between corresponding points in the source point cloud and the target point cloud in the first particle swarm; The weighted Cauchy kernel function is used to assign a first weight to the target point cloud corresponding points whose residuals are in the first threshold interval, and a second weight to the target point cloud corresponding points whose residuals are in the second threshold interval; wherein, the minimum residual value in the first threshold interval is greater than the maximum residual value in the second threshold interval, and the first weight is less than the second weight.

[0010] Based on the above technical solution, preferably, after obtaining the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm during each coarse registration and fine registration process, the method further includes: If the second fitness value is not less than the first fitness value, the PSO particle swarm optimization algorithm is used to determine the individual optimal position and the global optimal position, and the registration result is determined as the target registration result for the current round.

[0011] Based on the above technical solution, preferably, after obtaining the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm during each coarse registration and fine registration process, the method further includes: After completing a preset number of coarse registration rounds, the particles in the first particle swarm that has completed coarse registration are filtered based on their fitness values ​​to obtain an intermediate particle swarm with a preset ratio; the fitness value of the intermediate particle swarm is less than the fitness values ​​of other particles in the first particle swarm. The particles in the intermediate particle swarm are precisely registered using the ICP algorithm to obtain the second particle swarm.

[0012] More preferably, a second aspect of the present invention provides a point cloud registration system for power battery reconstruction, comprising: a relation acquisition module, a position registration module, and a pose optimization module; wherein, The relationship acquisition module is configured to establish an initial point correspondence between the source point cloud and the target point cloud by utilizing the geometric relationship between the feature descriptors constructed by pasting marker points on the surface of the battery to be registered; The location registration module is configured to use the PSO particle swarm optimization algorithm to simulate the correspondence of the initial points, update the position based on the individual optimal position and the global optimal position of the particle swarm, and obtain a first particle swarm after coarse registration; and combine the ICP algorithm to minimize the Euclidean distance between the corresponding points in the source point cloud and the target point cloud to determine the optimal rigid transformation, and obtain a second particle swarm after fine registration. The pose optimization module is configured to, during each coarse and fine registration process, acquire the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm. If the second fitness value is less than the first fitness value, fuse the second pose of the second particle swarm with the first pose of the first particle swarm to complete the correction of the first pose. Then, select the target pose that minimizes the fitness function value from the second pose as the global structure guide term to optimize the particle poses of the idle particle swarm of the first particle swarm to obtain the target particle swarm. The idle particle swarm includes particles that did not participate in the fine registration.

[0013] More preferably, a third aspect of the present invention provides an electronic device, including a processor and a memory; the memory has a computer program stored thereon, wherein the computer program, when executed by the processor, implements the point cloud registration method for power battery reconstruction described in the first aspect.

[0014] More preferably, a fourth aspect of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the point cloud registration method for power battery reconstruction described in the first aspect.

[0015] The point cloud registration method and system for power battery reconstruction of the present invention have the following advantages over the prior art: 1. By employing an elite-guided initialization strategy based on artificial markers, a high-quality starting point is provided for the global search. Building upon this, a master-slave collaborative optimization framework is constructed. Particle swarm optimization (PSO) is used for coarse registration in the global search, while an improved iterative nearest point algorithm (INP) is used for fine registration in the local optimization. By introducing a feedback adjustment mechanism and a global optimal guidance strategy, a bidirectional information exchange loop is established between PSO and ICP, enabling dynamic feedback of local optimization results to correct the global search direction, achieving collaborative evolution and significantly improving point cloud registration accuracy.

[0016] 2. In each coarse and fine registration process, the first fitness value and the second fitness value are compared. When the second fitness value is better, the second pose and the first pose are fused to correct the initial pose, generating a corrected pose that is closer to the real pose. This is used as the initial value for subsequent iterations, avoiding the direct transmission of coarse registration error to the fine registration stage. This forms a closed-loop optimization of "coarse-fine-recorrection", accurately guiding the particles to search for the optimal solution region, and significantly improving the convergence speed and overall registration accuracy of the algorithm.

[0017] 3. By selecting the target pose that minimizes the fitness function value from the second pose as the global structure guide term, the particle poses of the idle particle swarm in the first particle swarm are optimized. This allows particles that did not participate in the fine-tuning stage to perceive the local optimization trend, thereby realizing the global propagation of the fine registration optimization results throughout the entire particle swarm. This enhances the coupling and collaborative interaction between coarse and fine registration, reduces the information bias caused by focusing only on high-fitness particles and ignoring other particles during the fine registration optimization process, and maintains the dynamic balance between the overall evolutionary direction of the particle swarm and the local structure fitting ability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A schematic flowchart of a point cloud registration method for power battery reconstruction provided in an embodiment of the present invention; Figure 2 A schematic diagram illustrating the principle of the feature descriptor provided in this embodiment of the invention; Figure 3 This is a schematic diagram comparing the registration effects of different algorithms provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of four initial registration poses at 50%L provided in the embodiments of the present invention; Figure 5 A schematic diagram of the distance error distribution results based on color mapping provided in an embodiment of the present invention; Figure 6 A schematic diagram of a point cloud registration system for power battery reconstruction provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] In some embodiments, such as Figure 1 As shown, Figure 1 This is a flowchart illustrating a point cloud registration method for power battery reconstruction provided by an embodiment of the present invention; the point cloud registration method for power battery reconstruction provided by the present invention includes: S110, establish the initial point correspondence between the source point cloud and the target point cloud by utilizing the geometric relationship between the feature descriptors constructed by pasting marker points on the surface of the battery to be registered.

[0022] In this embodiment, prior geometric information is provided for point cloud registration by attaching marker points to the battery surface. These marker points can form a small but highly reliable set of correspondences. The initial pose is estimated based on the marker point correspondences, and initialization is completed, providing a high-quality search region for subsequent coarse registration. This effectively avoids problems such as slow convergence speed and instability caused by random initialization under large pose deviations.

[0023] In some embodiments, establishing an initial point correspondence between the source point cloud and the target point cloud using the geometric relationship between feature descriptors constructed by pasting marker points on the surface of the battery to be registered includes: Feature descriptions are performed on the marked points in the source point cloud and the target point cloud respectively, and the Euclidean distance between each feature vector is obtained. Using each marker in the source point cloud as a reference, search for the marker with the smallest Euclidean distance in the target point cloud to obtain the first matching set; and using each marker in the target point cloud as a reference, search for the marker with the smallest Euclidean distance in the source point cloud to obtain the second matching set. The initial point correspondence between the source point cloud and the target point cloud is established based on the intersection of the first matching set and the second matching set.

[0024] In this embodiment, please refer to Figure 2 , Figure 2 This is a schematic diagram illustrating the principle of the feature descriptor provided in an embodiment of the present invention; for a point cloud containing A marker points... Given one of the points And using this point as the center, use a KD-tree to quickly find its neighborhood and select the distance point. The most recent K points Calculation points To each of its K nearest neighbors Euclidean distance between The feature descriptors of the marked points are then constructed by arranging all the calculated Euclidean distances in ascending order. .

[0025] After performing feature descriptions on the marked points in the source point cloud S and the target point cloud Q respectively, a preliminary point-to-point relationship is established. For any point in the source point cloud S... By calculating its eigenvector With the eigenvectors of all points in Q The Euclidean distance between them measures their similarity, and is related to the points... matching points The point that minimizes this distance is determined as follows:

[0026] ; By traversing all marked points in the source point cloud S, we can obtain the initial set of candidate matching pairs: .

[0027] Considering the similarity of local features in point clouds and the multiple correspondence problem during point pair matching, erroneous matches may occur in the initial matching set. A bidirectional mutual selection principle is used to eliminate feature ambiguity and erroneous matches caused by one-to-many conflicts, thereby improving the matching quality. The specific implementation process is as follows:

[0028] Forward matching: Performs the nearest neighbor search from the source point cloud to the target point cloud, obtaining a matching set. ; Reverse matching: Perform a nearest neighbor search from the target point cloud to the source point cloud to obtain a matching set. ; Intersection filtering: The final reliable matching set is the intersection of the point clouds of the matching results in two directions. ; A filtering mechanism ensures that only matching pairs are retained. Simultaneously satisfy yes The nearest point in the target point cloud, and yes The closest point in the source point cloud.

[0029] In some embodiments, establishing an initial point correspondence between the source point cloud and the target point cloud based on the intersection of the first matching set and the second matching set includes: The intersection of point clouds is solved using the singular value decomposition method to obtain the optimal rigid body transformation parameters; The optimal rigid body transformation parameters are used as the initial elite particles, and multiple subsequent elite particles are generated by combining them with uniformly sampled random perturbation vectors. The initial point correspondence between the source point cloud and the target point cloud is established based on the initial elite particles and subsequent elite particles.

[0030] In this embodiment, the Singular Value Decomposition (SVD) method minimizes the distance error by decomposing the covariance matrix, transforming the solution of the rotation matrix into a combination problem of orthogonal matrices. The specific processing procedure for elite particles is as follows:

[0031] In a population of N particles, B elite particles are generated, among which... , To determine the proportion of elite particles, the particle positions are calculated from the marker points. As the first elite particle, the positions of the remaining B-1 elite particles are determined by... Add a uniformly sampled random perturbation vector to the top To generate, that is This ensures that elite particles are distributed within the neighborhood of high-quality solutions. The position of each particle... It is a six-dimensional rotation and translation vector , is used to represent a rigid body transformation. Rotational component disturbance range The translation component perturbation range is equal to the diagonal length of the point cloud bounding box. To ensure the global exploration capability of the particle swarm, the remaining... Each particle's position and velocity will be randomly initialized according to the standard PSO algorithm.

[0032] S120 uses the PSO (Particle Swarm Optimization) algorithm to simulate the correspondence of initial points, updates the particle swarm based on the individual optimal position and the global optimal position, and obtains the first particle swarm after coarse registration; combined with the ICP (Integrated Point Optimization) algorithm to minimize the Euclidean distance between corresponding points in the source point cloud and the target point cloud, the optimal rigid transformation is determined, and the second particle swarm after fine registration is obtained.

[0033] In some embodiments, the PSO (Particle Swarm Optimization) algorithm is used to simulate the initial point correspondence, and the initial point correspondence is updated based on the individual optimal positions and the global optimal positions of the particle swarm to obtain a first particle swarm after coarse registration, including: The current position of a particle is updated based on its historical velocity, historical position, historical best position of an individual particle, and historical best position of the entire swarm, to obtain the position with the minimum fitness value in this round. If the fitness value corresponding to the current position is less than the fitness value of the historical best position, replace the historical best position with the current position; otherwise, the historical best position remains unchanged.

[0034] In PSO, each candidate rigid body transformation is modeled as a particle, and its pose is iteratively updated through the interaction between the particle's own historical best state and the group's global best state. This collaborative update mechanism achieves a balance between global exploration and local exploitation. Its overall process can be summarized as follows:

[0035] Suppose there exists a particle swarm consisting of N particles in D-dimensional space. Each particle represents a rigid body transformation solution, where the i-th particle is... The position of this particle in D-dimensional space ,speed The optimal position of the i-th particle is... The optimal position fitness value obtained by this particle search is swarm global optimal position The optimal fitness value of the group is In the t-th iteration, the particle's velocity and position updates are expressed by the following equations:

[0036] ; ; in, It uses inertial weights, which have the ability to balance global and local search. and This is the velocity factor, used to control the flight speed of particles. and for Random numbers.

[0037] The formulas for updating the individual optimal position and the global optimal position are shown below: ; ; The particle position is updated by transforming the above two parameters.

[0038] In point cloud registration, the source point cloud S is aligned with the target point cloud Q through rigid body transformation to achieve spatial consistency. Ideally, each point... Its corresponding point The Euclidean distance between corresponding point pairs should be zero. Considering that noise, systematic errors, and incomplete scanning inevitably introduce biases in real-world measurement environments, the registration task can be formulated as a parametric optimization problem. The goal is to find the transformation that minimizes the overall distance between corresponding point pairs. To measure this difference, the root mean square error (RMSE) of the point pair distance is used as the optimization metric and forms the basis of the objective function defined by the following equation:

[0039] ; Where R and t represent the rotation matrix and translation vector, respectively, as defined in the following equation, and M represents the number of corresponding point pairs.

[0040] ; .

[0041] In some embodiments, before minimizing the Euclidean distance between corresponding points in the source point cloud and the target point cloud using the ICP algorithm, determining the optimal rigid transformation, and obtaining the finely registered second particle swarm, the method further includes: Obtain the residual between corresponding points in the source point cloud and the target point cloud in the first particle swarm; The weighted Cauchy kernel function is used to assign a first weight to the target point cloud corresponding points whose residuals are in the first threshold interval, and a second weight to the target point cloud corresponding points whose residuals are in the second threshold interval; wherein, the minimum residual value in the first threshold interval is greater than the maximum residual value in the second threshold interval, and the first weight is less than the second weight.

[0042] The core idea of ​​the ICP algorithm is to estimate the optimal rigid transformation by minimizing the Euclidean distance between corresponding points in the source and target point clouds, thereby achieving accurate spatial alignment. Specifically, for the source point cloud... Each point in In the target point cloud Find the corresponding target point in the middle Establish point-to-point sets Based on the established point correspondence, the rotation matrix and translation vector are updated by optimizing the objective function defined in the following formula.

[0043] ; Considering that the optimization process has not yet converged in the early stages of PSO iteration, the estimated pose may still deviate significantly from the true solution. Directly using such unstable poses as input to ICP would further increase the risk of premature or false convergence. Therefore, a weighted Cauchy kernel function is introduced to optimize the corresponding distance between point pairs. The weighted Cauchy kernel function assigns lower weights to corresponding points with larger residuals and higher weights to corresponding points with smaller residuals, thereby enhancing the influence of small-error corresponding points in the optimization process. This strategy effectively reduces the interference of outliers and improves the robustness and local geometric fitting accuracy of the algorithm. Cauchy weights The definition is as follows:

[0044] ; In the weighted model, Let represent the Euclidean residual of the i-th corresponding point pair, and the parameter c is used to control the suppression strength of the kernel function for large residuals. At that time, weight A value close to 1 indicates that the point corresponding to the small residual still has a strong influence during the optimization process; conversely, when... When the weights rapidly approach zero, the influence of outlier correspondences on rigid transformation estimation is effectively reduced. The scaling parameter c reflects the strength of outlier suppression; a larger c weakens the suppression effect, while a smaller c may lead to oversuppression. Here, the median absolute deviation (MAD) is used to determine the parameter c, and its calculation method is as follows:

[0045] ; The objective function based on the weighted Cauchy kernel function is defined as follows: .

[0046] S130, during each coarse and fine registration process, the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm are obtained. If the second fitness value is less than the first fitness value, the second pose of the second particle swarm is fused with the first pose of the first particle swarm to complete the correction of the first pose. The target pose that minimizes the fitness function value is selected from the second pose as the global structure guidance term to optimize the particle poses of the idle particle swarm of the first particle swarm to obtain the target particle swarm. The idle particle swarm includes particles that did not participate in fine registration.

[0047] In this embodiment, a unified evaluation index is constructed. The registration results of coarse and fine registration are quantitatively compared. If the ICP output is better than the PSO, the pose estimated by ICP is used to guide the particle to update the search direction, and the corresponding fitness value is reconstructed; otherwise, the coarse registration maintains the original update strategy and continues to perform the next iteration under the PSO framework. The algorithm terminates when the number of coarse registration iterations reaches a preset number, such as 100, or when the fitness value is less than a first preset threshold, such as 1, and the change in fitness value in consecutive iterations does not exceed a second preset threshold, such as 0.001.

[0048] In each iteration, the fitness value of the finely registered particles is recalculated to obtain... This fitness value is consistent with that in coarse registration. Maintaining consistency in evaluation provides a unified basis for judging the guiding effect. When the optimization results of the fine registration task are better than those of the coarse registration task, it satisfies the requirement... At that time, the refined pose output by the fine registration task will be... Current pose of the corresponding particle in the coarse registration task The particle poses are fused to correct the particle poses. The corrected poses serve as the initial state for the next iteration, guiding the subsequent search direction. The specific pose correction method is shown in the following equation:

[0049] ; in, This represents the transformation matrix calculated using the ICP algorithm. The six-dimensional pose parameters obtained after inversion. Coefficients. This is the particle correction factor, used to control the amplitude of pose correction. .

[0050] Considering the fine-registration task triggering mechanism, only the particles with the highest fitness values ​​will be further optimized. Therefore, most particles in the swarm cannot receive sufficient guidance during iteration, easily leading to an "island effect." Therefore, a globally optimal structure guiding term is introduced. This enables collaborative guidance and direction refinement at the group level. Specifically, the set of finely matched poses obtained from the task in the current iteration is denoted as... The pose that minimizes the fitness function value is selected as the global structure guide term. This vector is then embedded into the velocity update rule of the coarse registration task, resulting in the updated velocity expression:

[0051] ; in, For learning factors related to structure-guided terms, These are uniformly distributed random numbers.

[0052] By enabling particles that did not participate in the fine registration stage to perceive local optimization trends, the global propagation of the task optimization results throughout the entire particle swarm is achieved. This enhances the coupling and collaborative interaction between the coarse registration task and the fine registration task, reduces the information bias caused by focusing only on high-fitness particles and ignoring other particles during the fine registration optimization process, and thus maintains the dynamic balance between the overall evolutionary direction of the particle swarm and the local structure fitting ability.

[0053] In some embodiments, after obtaining the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm during each coarse registration and fine registration process, the method further includes: If the second fitness value is not less than the first fitness value, the PSO particle swarm optimization algorithm is used to determine the optimal position of the individual and the global optimal position, and the registration result is determined as the target registration result of the current round.

[0054] Here, when the optimization result of the coarse registration task is better than that of the fine registration task, it satisfies... If the registration result from the fine registration task is not used to guide the particle's search direction, then the coarse registration task continues to execute the optimization process independently according to the original update rules.

[0055] In some embodiments, after obtaining the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm during each coarse registration and fine registration process, the method further includes: After completing the coarse registration in a preset number of rounds, the particles in the first particle swarm that has completed coarse registration are filtered based on their fitness values ​​to obtain an intermediate particle swarm with a preset ratio; the fitness values ​​of the intermediate particle swarm are less than the fitness values ​​of other particles in the first particle swarm. The second particle swarm is obtained by finely registering the particles in the intermediate particle swarm based on the ICP algorithm.

[0056] In this embodiment, to achieve a balance between optimization accuracy and computational efficiency, a periodic fine-registration triggering strategy is adopted. Specifically, the fine-registration task is triggered every [period]. The coarse registration task is activated once, and a lightweight fine registration optimization is performed only on the top 10% of particles in terms of fitness from the coarse registration task. By comparing the fitness values ​​obtained from the coarse and fine registration tasks, it is determined whether the current particle needs pose correction.

[0057] In an optional embodiment, to quantitatively evaluate the registration performance of the proposed method, Mean Absolute Distance (MAD) and Root Mean Square Error (RMSE) are used as the main evaluation metrics, and their definitions and meanings are as follows: MAD (Modular Displacement) measures the average spatial deviation between the reconstructed point cloud and the reference model surface. It is expressed as the average of the absolute values ​​of the Euclidean distances from each reconstructed point to the ideal model surface. This metric is relatively insensitive to outliers and reflects the overall balance of registration accuracy. A smaller MAD value indicates higher registration accuracy. Its definition is as follows:

[0058] ; in, This indicates the number of point clouds after the device's 3D reconstruction. Let represent the i-th point in the reconstructed point cloud, and M represent the standard 3D model of the battery.

[0059] RMSE is used to quantify the average deviation between corresponding point pairs after registration. It is calculated as the square root of the mean of the squared Euclidean distances between matching point pairs in the source and target point clouds. RMSE comprehensively reflects the overall registration accuracy; the smaller the value, the higher the spatial alignment between the two point clouds. The definition of this metric is as follows:

[0060] ; in, This indicates the number of point pairs after point cloud registration. This represents the i-th point in the source point cloud. Indicates the relationship with the source point cloud The corresponding point.

[0061] ICP, PSO, S4PCS (four-point consistency set), PCA (principal component analysis) + ICP, and PSO + ICP were used as comparison methods to evaluate the registration effect of this application. To ensure fairness in the comparison, the relative pose parameters of the source and target point clouds were the same for all methods in the fair comparison experiment. The parameters of the comparison experiment were set according to typical experiments in various literatures. The comparison results are as follows: Figure 3 As shown in Table 1, Figure 3 This is a schematic diagram comparing the registration effects of different algorithms provided in embodiments of the present invention. Figure 3 As can be seen from the comparison of the results and the quantitative data in Table 1, the MTCOPCR method of this application shows a significant advantage in registration accuracy compared with the segmented registration method PSO+ICP and other traditional methods.

[0062] Table 1 Comparison of registration performance parameters of different algorithms To further evaluate the registration performance of this application compared to other algorithms under different initial pose deviations, an experimental dataset was constructed. Various initial poses were generated by applying different rotation and translation transformations to the source point cloud. Four rotation-based pose deviation scenarios were set up: from pose 1 to pose 4, with random rotation angles sampled from the intervals {[0°, 20°), [20°, 40°), [40°, 60°), [60°, 80°)}. For each rotation scenario, registration experiments were conducted with translation distances of {0, 25% × L, 50% × L, and 100% × L}, where L represents the diagonal length of the point cloud bounding box. See here for more details. Figure 4 , Figure 4 The following is a schematic diagram of four initial registration poses at 50%L provided in the embodiments of the present invention, showing the spatial distribution of the four rotation poses when the translation distance is fixed at 50%L.

[0063] based on Figure 4 The proposed method was compared with other staged registration methods, including PCA+ICP and PSO+ICP, using the pose configurations shown. The quantitative experimental results are summarized in Table 2. Table 2 shows that the proposed method achieved the highest registration accuracy in all four pose scenarios. Compared with PCA+ICP and PSO+ICP, the proposed method achieves higher registration accuracy without significantly sacrificing computational efficiency.

[0064] Table 2 Comparison of registration performance results under different poses and algorithms The reconstructed 3D models obtained using different registration methods were compared with the measurement results from a coordinate measuring machine (CMM) to evaluate the practical applicability of the reconstructed models in industrial inspection. The nearest point distance was used as the evaluation metric to quantify the overall deviation between the reconstructed point cloud and the standard 3D model. To visually represent the deviation distribution, a blue-red color mapping was used, and all marker points were removed before analysis to avoid bias in the results. Please refer to [link / reference]. Figure 5 , Figure 5This diagram illustrates the distance error distribution based on color mapping provided in this embodiment of the invention. The MTCOPCR method outperforms the traditional PSO+ICP method in overall performance. The MAD of the PSO+ICP method is 0.017 mm, while the proposed MTCOPCR method reduces the MAD to 0.013 mm, improving accuracy by approximately 23.5%. This result demonstrates that, under the same experimental conditions, the proposed method can effectively reduce the overall reconstruction error. By integrating residual feedback and structure-guided closed-loop optimization mechanisms, the fine registration process can dynamically correct the search direction in the coarse registration stage, thereby suppressing error accumulation and propagation, making the reconstruction results closer to the CMM standard model, and providing a reliable data foundation for subsequent geometric analysis and defect detection.

[0065] In some embodiments, please refer to Figure 6 , Figure 6 This is a schematic diagram of a point cloud registration system for power battery reconstruction provided in an embodiment of the present invention. The present invention provides a point cloud registration system 600 for power battery reconstruction, comprising: a relation acquisition module 610, a position registration module 620, and a pose optimization module 630; wherein,

[0066] The relationship acquisition module 610 is configured to establish an initial point correspondence between the source point cloud and the target point cloud by utilizing the geometric relationship between the feature descriptors constructed by pasting marker points on the surface of the battery to be registered; The location registration module 620 is configured to use the PSO particle swarm optimization algorithm to simulate the correspondence of the initial points, update the position based on the individual optimal position and the global optimal position of the particle swarm, and obtain the first particle swarm after coarse registration; combine the ICP algorithm to minimize the Euclidean distance between the corresponding points in the source point cloud and the target point cloud, determine the optimal rigid transformation, and obtain the second particle swarm after fine registration. The pose optimization module 630 is configured to, during each coarse and fine registration process, acquire the second fitness value corresponding to the second particle group and the first fitness value corresponding to the first particle group. If the second fitness value is less than the first fitness value, the second pose of the second particle group is fused with the first pose of the first particle group to complete the correction of the first pose. The target pose that minimizes the fitness function value is selected from the second pose as the global structure guidance term to optimize the particle poses of the idle particle group of the first particle group to obtain the target particle group. The idle particle group includes particles that did not participate in fine registration.

[0067] In some embodiments, the relationship acquisition module 610 is specifically configured as follows: Feature descriptions are performed on the marked points in the source point cloud and the target point cloud respectively, and the Euclidean distance between each feature vector is obtained. Using each marker in the source point cloud as a reference, search for the marker with the smallest Euclidean distance in the target point cloud to obtain the first matching set; and using each marker in the target point cloud as a reference, search for the marker with the smallest Euclidean distance in the source point cloud to obtain the second matching set. The initial point correspondence between the source point cloud and the target point cloud is established based on the intersection of the first matching set and the second matching set.

[0068] In some embodiments, the relationship acquisition module 610 is specifically configured as follows: The intersection of point clouds is solved using the singular value decomposition method to obtain the optimal rigid body transformation parameters; The optimal rigid body transformation parameters are used as the initial elite particles, and multiple subsequent elite particles are generated by combining them with uniformly sampled random perturbation vectors. The initial point correspondence between the source point cloud and the target point cloud is established based on the initial elite particles and subsequent elite particles.

[0069] In some embodiments, the location registration module 620 is specifically configured as follows: The current position of a particle is updated based on its historical velocity, historical position, historical best position of an individual particle, and historical best position of the entire swarm, to obtain the position with the minimum fitness value in this round. If the fitness value corresponding to the current position is less than the fitness value of the historical best position, replace the historical best position with the current position; otherwise, the historical best position remains unchanged.

[0070] In some embodiments, the location registration module 620 is further configured to: Obtain the residual between corresponding points in the source point cloud and the target point cloud in the first particle swarm; The weighted Cauchy kernel function is used to assign a first weight to the target point cloud corresponding points whose residuals are in the first threshold interval, and a second weight to the target point cloud corresponding points whose residuals are in the second threshold interval; wherein, the minimum residual value in the first threshold interval is greater than the maximum residual value in the second threshold interval, and the first weight is less than the second weight.

[0071] In some embodiments, the pose optimization module 630 is specifically configured as follows: If the second fitness value is not less than the first fitness value, the PSO particle swarm optimization algorithm is used to determine the optimal position of the individual and the global optimal position, and the registration result is determined as the target registration result of the current round.

[0072] In some embodiments, the pose optimization module 630 is specifically configured as follows: After completing the coarse registration in a preset number of rounds, the particles in the first particle swarm that has completed coarse registration are filtered based on their fitness values ​​to obtain an intermediate particle swarm with a preset ratio; the fitness values ​​of the intermediate particle swarm are less than the fitness values ​​of other particles in the first particle swarm. The second particle swarm is obtained by finely registering the particles in the intermediate particle swarm based on the ICP algorithm.

[0073] It should be noted that the point cloud registration system for power battery reconstruction provided in this application embodiment and the point cloud registration method for power battery reconstruction provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned point cloud registration method for power battery reconstruction, and the repeated parts will not be described again.

[0074] In some embodiments, please refer to Figure 7 , Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 700 provided in this application includes a processor 710 and a memory 720; the memory 720 stores a computer program, wherein the computer program, when executed by the processor, implements the aforementioned point cloud registration method for power battery reconstruction.

[0075] Specifically, processor 710 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. Processor 710 may also include onboard memory for caching purposes. Processor 710 may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0076] The memory 720 can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, the memory 720 can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, apparatus, or propagation medium. Specific examples of the memory 720 include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0077] This application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the point cloud registration method for power battery reconstruction described above. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0078] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0079] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A point cloud registration method for power battery reconstruction, characterized in that, include: The initial point correspondence between the source point cloud and the target point cloud is established by utilizing the geometric relationship between the feature descriptors constructed by pasting marker points on the surface of the battery to be registered; The PSO (Particle Swarm Optimization) algorithm is used to simulate the correspondence of the initial points. The particle swarm is updated based on the individual optimal position and the global optimal position to obtain the first particle swarm after coarse registration. The ICP (Integrated Point Plane) algorithm is combined to minimize the Euclidean distance between the corresponding points in the source point cloud and the target point cloud to determine the optimal rigid transformation and obtain the second particle swarm after fine registration. In each coarse and fine registration process, the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm are obtained. If the second fitness value is less than the first fitness value, the second pose of the second particle swarm is fused with the first pose of the first particle swarm to complete the correction of the first pose. The target pose that minimizes the fitness function value is selected from the second pose as the global structure guide term to optimize the particle poses of the idle particle swarm of the first particle swarm to obtain the target particle swarm. The idle particle swarm includes particles that did not participate in fine registration.

2. The point cloud registration method for power battery reconstruction as described in claim 1, characterized in that, The step of establishing the initial point correspondence between the source point cloud and the target point cloud by utilizing the geometric relationship between the feature descriptors constructed by pasting marker points on the surface of the battery to be registered includes: The marked points in the source point cloud and the target point cloud are described by features respectively, and the Euclidean distance between each feature vector is obtained. Using each marker point in the source point cloud as a reference, search for the marker point with the smallest Euclidean distance in the target point cloud to obtain a first matching set; and using each marker point in the target point cloud as a reference, search for the marker point with the smallest Euclidean distance in the source point cloud to obtain a second matching set. The initial point correspondence between the source point cloud and the target point cloud is established based on the intersection of the point clouds of the first matching set and the second matching set.

3. The point cloud registration method for power battery reconstruction as described in claim 2, characterized in that, The step of establishing the initial point correspondence between the source point cloud and the target point cloud based on the intersection of the first matching set and the second matching set includes: The intersection of the point cloud is solved using the singular value decomposition method to obtain the optimal rigid body transformation parameters; The optimal rigid body transformation parameters are used as initial elite particles, and multiple subsequent elite particles are generated by combining them with uniformly sampled random perturbation vectors. The initial point correspondence between the source point cloud and the target point cloud is established based on the initial elite particles and the subsequent elite particles.

4. The point cloud registration method for power battery reconstruction as described in claim 1, characterized in that, The initial point correspondence is simulated using the PSO particle swarm optimization algorithm, and updated based on the individual optimal positions and global optimal positions of the particle swarm to obtain a first particle swarm after coarse registration, including: The current position of a particle is updated based on its historical velocity, historical position, historical individual best position, and historical group best position, to obtain the position with the minimum fitness value in this round. If the fitness value corresponding to the current position is less than the fitness value of the historical individual's best position, the current position replaces the historical individual's best position; otherwise, the historical individual's best position remains unchanged.

5. The point cloud registration method for power battery reconstruction as described in claim 1, characterized in that, Before combining the ICP algorithm to minimize the Euclidean distance between corresponding points in the source and target point clouds, determine the optimal rigid transformation, and obtain the finely registered second particle swarm, the process also includes: Obtain the residual between corresponding points in the source point cloud and the target point cloud in the first particle swarm; The weighted Cauchy kernel function is used to assign a first weight to the target point cloud corresponding points whose residuals are in the first threshold interval, and a second weight to the target point cloud corresponding points whose residuals are in the second threshold interval; wherein, the minimum residual value in the first threshold interval is greater than the maximum residual value in the second threshold interval, and the first weight is less than the second weight.

6. The point cloud registration method for power battery reconstruction as described in claim 1, characterized in that, After obtaining the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm during each coarse registration and fine registration process, the method further includes: If the second fitness value is not less than the first fitness value, the PSO particle swarm optimization algorithm is used to determine the individual optimal position and the global optimal position, and the registration result is determined as the target registration result for the current round.

7. The point cloud registration method for power battery reconstruction as described in claim 1, characterized in that, After obtaining the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm during each coarse registration and fine registration process, the method further includes: After completing a preset number of coarse registration rounds, the particles in the first particle swarm that has completed coarse registration are filtered based on their fitness values ​​to obtain an intermediate particle swarm with a preset ratio; the fitness value of the intermediate particle swarm is less than the fitness values ​​of other particles in the first particle swarm. The particles in the intermediate particle swarm are precisely registered using the ICP algorithm to obtain the second particle swarm.

8. A point cloud registration system for power battery reconstruction, characterized in that, include: The module includes a relationship acquisition module, a position registration module, and a pose optimization module; among them, The relationship acquisition module is configured to establish an initial point correspondence between the source point cloud and the target point cloud by utilizing the geometric relationship between the feature descriptors constructed by pasting marker points on the surface of the battery to be registered; The location registration module is configured to use the PSO particle swarm optimization algorithm to simulate the correspondence of the initial points, update the position based on the individual optimal position and the global optimal position of the particle swarm, and obtain a first particle swarm after coarse registration; and combine the ICP algorithm to minimize the Euclidean distance between the corresponding points in the source point cloud and the target point cloud to determine the optimal rigid transformation, and obtain a second particle swarm after fine registration. The pose optimization module is configured to, during each coarse and fine registration process, acquire the second fitness value corresponding to the second particle swarm and the first fitness value corresponding to the first particle swarm. If the second fitness value is less than the first fitness value, fuse the second pose of the second particle swarm with the first pose of the first particle swarm to complete the correction of the first pose. Then, select the target pose that minimizes the fitness function value from the second pose as the global structure guide term to optimize the particle poses of the idle particle swarm of the first particle swarm to obtain the target particle swarm. The idle particle swarm includes particles that did not participate in the fine registration.

9. An electronic device comprising a processor and a memory; said memory storing a computer program, wherein, When the computer program is executed by the processor, it implements the point cloud registration method for power battery reconstruction as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, It stores a computer program, wherein the computer program, when executed by a processor, implements the point cloud registration method for power battery reconstruction as described in any one of claims 1 to 7.

Images