Three-dimensional reconstruction method and system of crime scene based on three-dimensional laser

Abstract

The embodiment of the present invention provides a method and system for 3D reconstruction of a crime scene based on 3D laser, which belongs to the technical field of 3D reconstruction and case investigation. The method includes: determining a target crime scene, and using a target sensor to collect data from the target crime scene to obtain initial point cloud data; performing data filtering processing on the initial point cloud data to obtain target point cloud data; using a density-based clustering algorithm to perform data segmentation on the target point cloud data to obtain a first segmentation result; using a growth direction and color fusion strategy to perform data segmentation on the target point cloud data to obtain a second segmentation result; fusing the first segmentation result and the second segmentation result to obtain a target segmentation result corresponding to the target point cloud data; and determining a 3D reconstruction result corresponding to the target crime scene according to the target segmentation result.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional reconstruction and case investigation, and in particular to a three-dimensional reconstruction method and system for a crime scene based on three-dimensional laser. Background Art

[0002] At the scene of the incident, using 3D reconstruction technology for on-site simulation is an efficient method that can provide users with detailed reference information, thereby significantly improving the efficiency of problem solving. 3D reconstruction technology can reconstruct a virtual model close to the actual scene by capturing detailed information on the scene, including the shape, position and environmental conditions of objects. This model can not only help users better understand the layout and details of the scene, but also provide strong support in subsequent analysis and decision-making. However, although 3D reconstruction technology has shown its great potential in many fields, the related technologies have the problem of being unable to accurately perform point cloud cutting during the 3D reconstruction process, resulting in low accuracy and precision of the reconstruction results. Summary of the invention

[0003] The main purpose of the embodiments of the present invention is to provide a method and system for three-dimensional reconstruction of crime scenes based on three-dimensional laser, aiming to solve the problem in the related art that point cloud cutting cannot be accurately performed during the three-dimensional reconstruction process, resulting in low accuracy and precision of the reconstruction results.

[0004] In a first aspect, an embodiment of the present invention provides a method for three-dimensional reconstruction of a crime scene based on three-dimensional laser, comprising:

[0005] Determine the target case event scene, and use the target sensor to collect data on the target case event scene to obtain initial point cloud data;

[0006] Performing data filtering processing on the initial point cloud data to obtain target point cloud data;

[0007] Using a density-based clustering algorithm to perform data segmentation on the target point cloud data to obtain a first segmentation result;

[0008] Performing data segmentation on the target point cloud data using a fusion strategy of growth direction and color to obtain a second segmentation result;

[0009] fusing the first segmentation result and the second segmentation result to obtain a target segmentation result corresponding to the target point cloud data;

[0010] A three-dimensional reconstruction result corresponding to the target case event scene is determined according to the target segmentation result.

[0011] In a second aspect, an embodiment of the present invention provides a three-dimensional laser-based crime scene three-dimensional reconstruction system, comprising:

[0012] A data acquisition module is used to determine the target case event scene and use a target sensor to collect data from the target case event scene to obtain initial point cloud data;

[0013] A data filtering module, used for performing data filtering processing on the initial point cloud data to obtain target point cloud data;

[0014] A first segmentation module, used to perform data segmentation on the target point cloud data using a density-based clustering algorithm to obtain a first segmentation result;

[0015] A second segmentation module, used for performing data segmentation on the target point cloud data by using a fusion strategy of growth direction and color to obtain a second segmentation result;

[0016] A data fusion module, used for fusing the first segmentation result and the second segmentation result to obtain a target segmentation result corresponding to the target point cloud data;

[0017] The three-dimensional reconstruction module is used to determine the three-dimensional reconstruction result corresponding to the target case event scene according to the target segmentation result.

[0018] The embodiment of the present invention provides a method and system for three-dimensional reconstruction of a crime scene based on three-dimensional laser. The method includes determining a target crime scene, and then using a target sensor to collect data on the target crime scene to obtain initial point cloud data, and then performing data filtering processing on the initial point cloud data to obtain target point cloud data, thereby effectively removing noise and unnecessary point cloud data, reducing the processing burden, and improving the efficiency and accuracy of subsequent processing steps. Thus, the density distribution of the target point cloud data is analyzed based on a density clustering algorithm, and different objects or regions can be effectively identified and segmented to obtain a first segmentation result, and the target point cloud data is segmented using a fusion strategy of growth direction and color to obtain a second segmentation result. Thus, through multi-strategy fusion, the advantages of the first segmentation result and the second segmentation result are combined, and a more accurate target segmentation result can be finally obtained. This accurate segmentation result provides clear object boundaries and structural information for three-dimensional reconstruction, making the reconstruction result more refined and realistic, and then the target crime scene is reconstructed in three dimensions according to the target segmentation result to obtain a corresponding three-dimensional reconstruction result. This process can effectively restore the virtual scene of the crime scene, helping investigators to more intuitively understand and analyze the situation at the crime scene. At the same time, by optimizing the quality of 3D reconstruction, the accuracy and reliability of the reconstruction results are further improved, providing important support for subsequent analysis and application. It also solves the problem in related technologies that the accuracy and precision of the reconstruction results are low due to the inability to accurately cut point clouds during 3D reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 A schematic flow chart of a method for three-dimensional reconstruction of a crime scene based on three-dimensional laser provided in an embodiment of the present invention;

[0021] Figure 2 A schematic diagram of the module structure of a three-dimensional laser-based crime scene three-dimensional reconstruction system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0024] It should be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0025] The embodiment of the present invention provides a method and system for 3D reconstruction of crime scene based on 3D laser. The method can be applied to a terminal device, which can be an electronic device such as a tablet computer, a laptop computer, a desktop computer, a personal digital assistant, and a wearable device. The terminal device can be a server or a server cluster.

[0026] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0027] Please refer to Figure 1 , Figure 1A schematic flow chart of a method for three-dimensional reconstruction of a crime scene based on three-dimensional laser provided in an embodiment of the present invention.

[0028] like Figure 1 As shown, the three-dimensional laser-based crime scene three-dimensional reconstruction method includes steps S101 to S106.

[0029] Step S101: determine the target case event scene, and use the target sensor to collect data on the target case event scene to obtain initial point cloud data.

[0030] Exemplarily, the target case event scene refers to the target case scene and the target event scene, so that a suitable target sensor is selected according to the complexity and required accuracy of the target case event scene. Common target sensors include but are not limited to laser scanners, RGB-D cameras, multispectral imagers, etc. Ensure that the selected target sensor can meet the needs of data collection, so as to obtain initial point cloud data by collecting data at the target case event scene according to the target sensor.

[0031] Step S102: performing data filtering processing on the initial point cloud data to obtain target point cloud data.

[0032] Exemplarily, a statistical filtering algorithm is used to remove isolated points and unreasonable points according to the statistical characteristics of the point cloud data in the initial point cloud data to obtain the target point cloud data.

[0033] In some embodiments, the data filtering processing of the initial point cloud data to obtain the target point cloud data includes: performing a neighbor search on the initial point cloud data to obtain the first associated neighbor data corresponding to each first sub-point cloud data in the initial point cloud data; calculating the first distance between the first sub-point cloud data and the first associated neighbor data, and determining the second distance and standard deviation value corresponding to the initial point cloud data according to the first distance; performing noise removal on the initial point cloud data according to the second distance and the standard deviation value to obtain denoised point cloud data; performing a neighbor search on the denoised point cloud data to obtain the second associated neighbor data corresponding to each second sub-point cloud data in the denoised point cloud data; performing center calculation according to the second associated neighbor data to obtain the target center corresponding to the second sub-point cloud data; performing linear fitting according to the second associated neighbor data and the target center to obtain the target straight line parameters corresponding to the second sub-point cloud data, the target straight line parameters including a slope coefficient and a constant coefficient; performing data filtering on the second sub-point cloud data according to the target straight line parameters to obtain the target point cloud data; wherein the slope coefficient and the constant coefficient are obtained according to the following formula:

[0034]

[0035]

[0036] in, represents the slope coefficient corresponding to the i-th second sub-point cloud data, represents the constant coefficient corresponding to the i-th second sub-point cloud data, n represents the number of neighbors corresponding to the second associated neighbor data, represents the distance between the i-th second sub-point cloud data and the j-th second associated neighbor data, represents the target center corresponding to the i-th second sub-point cloud data, represents the smoothing parameter.

[0037] Exemplarily, an efficient data structure such as a KD tree or an octree is used to perform a neighbor search on the initial point cloud data, and the neighbor data corresponding to each first sub-point cloud data in the initial point cloud data is found to obtain the first associated neighbor data.

[0038] Exemplarily, the first distance is obtained by calculating the distance between the first sub-point cloud data and the first associated neighbor data using the Euclidean distance. Thus, all first distances are summarized and analyzed according to the calculated first distance. The second distance is obtained by calculating the average value of all first distances in the initial point cloud data. And the standard deviation value of all first distances in the initial point cloud data is calculated. The standard deviation value reflects the degree of dispersion of the distance between the first sub-point cloud data and the first associated neighbor data in the initial point cloud data.

[0039] Exemplarily, a threshold for noise identification is set according to the second distance and the standard deviation value. The first sub-point cloud data whose first distance is significantly greater than the second distance and the standard deviation value is identified as a noise point, and then the identified noise point is removed from the initial point cloud data to obtain denoised point cloud data. This ensures that the denoised point cloud data is purer and reduces the computational burden of subsequent processing.

[0040] Exemplarily, the denoised point cloud data after denoising is subjected to a neighbor search using an efficient data structure such as a KD tree or an octree to obtain the neighbor data corresponding to each second sub-point cloud data in the denoised point cloud data to obtain the second associated neighbor data.

[0041] Exemplarily, the target center of each second sub-point cloud data is obtained by calculating the average coordinates of the second associated neighbor data, and then a linear fit is performed on each second sub-point cloud data and its corresponding second associated neighbor data using a linear regression method. During the fitting process, the best straight line equation is found to obtain the target straight line parameters, including the slope coefficient and the constant coefficient. The slope coefficient and the constant coefficient are obtained according to the following formula:

[0042]

[0043]

[0044] in, Represents the slope coefficient corresponding to the i-th second sub-point cloud data, represents the constant coefficient corresponding to the i-th second sub-point cloud data, n represents the number of neighbors corresponding to the second associated neighbor data, represents the distance between the i-th second sub-point cloud data and the j-th second associated neighbor data, Indicates the target center corresponding to the i-th second sub-point cloud data, represents the smoothing parameter.

[0045] Exemplarily, the slope coefficient and constant coefficient are obtained by linear fitting, so that the geometric features of the initial point cloud data can be maintained during the filtering process. The straight line equation obtained by linear fitting can ensure that the filtered point cloud data is corrected along the direction of the target straight line, retaining important geometric information. And the number of neighbors and the point cloud distance are comprehensively considered to ensure that each second sub-point cloud data is adjusted based on its nearby points during the filtering process, so that the consistency of the denoised point cloud data can be improved through the local adjustment strategy, making the filtered target point cloud data more uniform and coherent as a whole. In addition, by introducing the target center, it is ensured that the filtering process not only focuses on local geometric features, but also maintains the integrity of the overall target, thereby ensuring the accuracy of the filtering results.

[0046] Exemplarily, a fitting straight line corresponding to the second sub-point cloud data is obtained according to the calculated target straight line parameters, and the second sub-point cloud data is substituted into the fitting straight line to obtain a filtering result, thereby implementing data filtering of the second sub-point cloud data, and obtaining the corresponding target point cloud data.

[0047] Step S103: using a density-based clustering algorithm to perform data segmentation on the target point cloud data to obtain a first segmentation result.

[0048] Exemplarily, the neighborhood radius and minimum number of points of the target point cloud data are determined so as to use DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to traverse the target point cloud data for data clustering so that each sub-point cloud data is assigned to a cluster to form multiple clusters. Each cluster represents a segmented area in the point cloud data, thereby obtaining the first segmentation result corresponding to the target point cloud data.

[0049] In some embodiments, the use of a density-based clustering algorithm to perform data segmentation on the target point cloud data to obtain a first segmentation result includes: performing a neighbor search on the target point cloud data to obtain third associated neighbor data corresponding to each third sub-point cloud data in the target point cloud data; calculating a third distance between the third sub-point cloud data and each first sub-neighbor data in the third associated neighbor data; determining a neighbor radius corresponding to the third sub-point cloud data based on the third distance, and performing data clustering on the target point cloud data based on the neighbor radius and the third sub-point cloud data to obtain an initial clustering result; performing a mutually reachable density test on the fourth sub-point cloud data of each subclass cluster in the initial clustering result to obtain a target clustering result; and performing data segmentation on the target point cloud data based on the target clustering result to obtain the first segmentation result.

[0050] Exemplarily, a K-nearest neighbor (KNN) or a grid-based neighbor search algorithm is used to perform a neighbor search on the target point cloud data to obtain third associated neighbor data corresponding to each third sub-point cloud data in the target point cloud data.

[0051] Exemplarily, a distance measurement algorithm such as Euclidean distance and Manhattan distance is used to calculate the third distance between each third sub-point cloud data and the third associated neighbor data, and then the neighbor radius corresponding to each third sub-point cloud data is determined based on the third distance using average distance, median distance or other statistical methods.

[0052] For example, the target point cloud data is clustered using the determined neighbor radius according to the clustering algorithm DBSCAN to obtain the initial clustering result, and then the fourth sub-point cloud data of each sub-cluster in the initial clustering result is tested for mutual reachability density. That is, it is checked whether each point is within the neighbor radius of other points to ensure density reachability, and then the sub-cluster is adjusted or merged according to the reachability test result to obtain the target clustering result.

[0053] Exemplarily, the segmentation criterion is determined according to the target clustering result, and each sub-class cluster represents a segmentation region. Thus, the target point cloud data is segmented according to the segmentation criterion to obtain the first segmentation result. Each segmentation region corresponds to the point cloud data of a sub-class cluster.

[0054] In some embodiments, determining the neighbor radius corresponding to the third sub-point cloud data according to the third distance includes: calculating the first distance difference between the adjacent first sub-neighbor data in the third associated neighbor data according to the third distance; performing data screening on the third associated neighbor data according to the first distance difference and a preset threshold to obtain fourth associated neighbor data; obtaining a fourth distance between the fourth associated neighbor data and the third sub-point cloud data from the third distance; calculating the second distance difference between the adjacent second sub-neighbor data in the fourth associated neighbor data according to the fourth distance; determining the neighbor radius corresponding to the third sub-point cloud data according to the second distance difference and the fourth distance; wherein the neighbor radius is obtained according to the following formula:

[0055]

[0056] in, represents the neighbor radius corresponding to the i-th third sub-point cloud data, s represents the number of neighbors corresponding to the fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the j-th fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the j-1-th fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the first fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the s-th fourth associated neighbor data, Represents the second distance difference between adjacent second sub-neighborhood data.

[0057] Exemplarily, in the third associated neighbor data, adjacent first sub-neighbor data are determined. The adjacent first sub-neighbor data are composed of points that are sequentially adjacent in the third associated neighbor data. For each pair of adjacent first sub-neighbor data, the difference of the third distance between them is calculated to obtain the first distance difference. That is, it is completed by taking the third distance of two adjacent first sub-neighbor data and then subtracting them.

[0058] Exemplarily, a preset threshold is set for screening the third associated neighbor data. This threshold can be determined according to the specific application scenario and data characteristics. The first distance difference is compared with the preset threshold. If the first distance difference is less than or equal to the preset threshold, the corresponding first sub-neighbor data is retained; otherwise, these data are eliminated to obtain the screened fourth associated neighbor data.

[0059] Exemplarily, distance values ​​corresponding to the fourth associated neighbor data are extracted from the previously calculated third distance, and these distance values ​​represent the distance between the fourth associated neighbor data and the third sub-point cloud data, that is, the fourth distance.

[0060] Exemplarily, adjacent second sub-neighborhood data are determined in the fourth associated neighbor data. The adjacent second sub-neighborhood data are sequentially adjacent to the determined adjacent points, and then for each pair of adjacent second sub-neighborhood data, the difference in the fourth distance between them is calculated to obtain the second distance difference. Thus, the neighbor radius corresponding to the third sub-point cloud data is determined using the second distance difference and the fourth distance according to the following formula:

[0061]

[0062] in, represents the neighbor radius corresponding to the i-th third sub-point cloud data, s represents the number of neighbors corresponding to the fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the j-th fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the j-1-th fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the first fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the s-th fourth associated neighbor data, Represents the second distance difference between adjacent second sub-neighborhood data.

[0063] For example, by calculating the second distance difference between adjacent second sub-neighborhood data, the local density changes of the point cloud data can be more accurately captured. This precise local information is crucial for density-based clustering algorithms because it can help the algorithm more accurately identify areas of different densities, thereby improving the accuracy of clustering. Determining the neighbor radius corresponding to the third sub-point cloud data according to the above formula can make the neighbor radius more adaptive so that the neighbor radius can be automatically adjusted according to the data density of different regions, thereby better adapting to the local characteristics of the data. In addition, since the distance difference between adjacent points is calculated, it is easier to identify points that are significantly different from the surrounding points, so that they are regarded as noise or outliers and processed, and then by accurately calculating the neighbor radius of each third sub-point cloud data, the detailed characteristics of the data can be better preserved.

[0064] Step S104: segment the target point cloud data using a fusion strategy of growth direction and color to obtain a second segmentation result.

[0065] Exemplarily, the color information of each point cloud data is extracted from the target point cloud data, and the growth direction of each point cloud data is calculated based on the normal vector, gradient vector, etc. The weight of the growth direction and color information in the segmentation process is determined using a weight coefficient. Thus, the growth direction and color information are combined according to the weight coefficient. For example, the vector of the growth direction and the color information can be weighted and summed to obtain a comprehensive feature vector.

[0066] Exemplarily, a set of initial seed points are selected from the target point cloud data. These seed points can be random or selected based on some preset rules (such as density, color, growth direction, etc.). For each seed point, a neighborhood search is performed to find its neighboring points. The definition of neighboring points can be determined based on a distance threshold or a spatial relationship. For each neighboring point, its comprehensive feature vector is calculated and compared with the comprehensive feature vector of the seed point based on metrics such as Euclidean distance and cosine similarity. If the comprehensive feature vector of the neighboring point is similar to the comprehensive feature vector of the seed point within a certain threshold, the neighboring point is merged into the current area and used as a new seed point for the next round of growth.

[0067] Exemplarily, the neighborhood search and feature comparison are repeated until no new points can be merged into the current region. Then, a new seed point is selected and a new growth process is started until all point cloud data are assigned to a region, so that after each point cloud data is assigned to a region, multiple sub-regions are formed. Each sub-region represents a segmentation result, thereby obtaining a second segmentation result.

[0068] Exemplarily, the target point cloud data is segmented using a fusion strategy of growth direction and color information to obtain a second segmentation result. This fusion strategy combines the geometric information of the growth direction and the visual information of the color, and can more comprehensively reflect the characteristics of the point cloud data, thereby improving the accuracy and robustness of the segmentation.

[0069] In some embodiments, the data segmentation of the target point cloud data using a fusion strategy of growth direction and color to obtain a second segmentation result includes: voxelizing the target point cloud data to obtain super voxel data, and determining a seed voxel and adjacent voxels corresponding to the seed voxel based on the super voxel data; calculating a normal angle and a color difference between the seed voxel and the adjacent voxels; merging the seed voxel and the adjacent voxel based on the normal angle and the color difference to obtain a regional growth direction corresponding to the seed voxel; regionally segmenting the super voxel data based on the regional growth direction and the seed voxel to obtain a voxel segmentation result; and determining the second segmentation result corresponding to the target point cloud data based on the voxel segmentation result.

[0070] Exemplarily, a suitable voxel size is selected to ensure that the target point cloud data can be evenly divided into each voxel. The target point cloud data is divided into a plurality of voxels, each voxel containing a group of spatially adjacent point cloud data to obtain supervoxel data.

[0071] Exemplarily, a group of seed voxels are selected from the voxelized supervoxel data. The seed voxels may be random or may be selected based on a certain preset rule (such as density, color, spatial position, etc.).

[0072] For each seed voxel, its neighboring voxels are determined. The definition of neighboring voxels can be voxels that share faces, edges or vertices with the seed voxel.

[0073] Exemplarily, the normal vectors of each seed voxel and its neighboring voxels are calculated by local smoothing or PCA (principal component analysis) method, thereby calculating the normal angle between the seed voxel and its neighboring voxels. The calculation of the normal angle can be achieved by the dot product or cross product of two normal vectors. And the color difference between the seed voxel and its neighboring voxels is calculated by calculating the Euclidean distance of the midpoint colors of the two voxels.

[0074] For example, if the normal angle is less than a preset angle and the color difference is less than a preset difference, the seed voxel and the adjacent voxel are merged to form a larger region, and during the merging process, the normal direction and color characteristics of each region are recorded, and a comprehensive regional growth direction is calculated. This direction can be used as a reference for subsequent regional segmentation. If the normal angle is greater than or equal to the preset angle or the color difference is greater than or equal to the preset difference, the seed voxel and the adjacent voxel are not merged.

[0075] Exemplarily, the neighborhood search and feature comparison are continued according to the region growth direction, and the adjacent voxels that meet the conditions are gradually merged into the current region. This process can be iterated until there are no new voxels to be merged. Each region represents a segmentation result to form a voxel segmentation result. The segmentation result can be represented by marking the region to which each voxel belongs. Thus, the voxel segmentation result is mapped back to the original target point cloud data, and then the second segmentation result corresponding to the target point cloud data is obtained according to the voxel segmentation result.

[0076] Step S105 : fusing the first segmentation result and the second segmentation result to obtain a target segmentation result corresponding to the target point cloud data.

[0077] Exemplarily, a common reference point or reference region is selected to ensure that the first segmentation result and the second segmentation result are aligned at the same spatial position and scale. If there is a difference in the coordinate system between the first segmentation result and the second segmentation result, necessary coordinate transformation is performed to align them in the same spatial coordinate system.

[0078] Exemplarily, the features of each region are extracted from the first segmentation result, such as color, normal, shape, etc. The features of each region are extracted from the second segmentation result, such as color difference, normal angle, local density, etc. For each region, the feature similarity between the corresponding regions in the first segmentation result and the second segmentation result is calculated. The similarity can be calculated by measurement methods such as Euclidean distance, cosine similarity, mutual information, etc., so as to determine the corresponding region of each region in the first segmentation result and the second segmentation result according to the feature similarity. Perform region matching to ensure that each first segmentation region finds its corresponding second segmentation region. Thus, the first segmentation region and the second segmentation region are data fused to obtain the target segmentation result corresponding to the target point cloud data.

[0079] In some embodiments, the fusion of the first segmentation result and the second segmentation result to obtain the target segmentation result corresponding to the target point cloud data includes: performing data merging on the first segmentation result and the second segmentation result to obtain an initial segmentation result; performing data classification analysis on the initial segmentation result to obtain the target segmentation result corresponding to the target point cloud data.

[0080] Exemplarily, the features of each region are extracted from the first segmentation result, such as color, normal, shape, etc. The features of each region are extracted from the second segmentation result, such as color difference, normal angle, local density, etc. For each region, the feature similarity between the corresponding regions in the first segmentation result and the second segmentation result is calculated. The similarity can be calculated by metrics such as Euclidean distance, cosine similarity, mutual information, etc., so as to determine the corresponding region in each region in the first segmentation result and the second segmentation result based on the feature similarity. Perform region matching to ensure that each first segmentation region finds its corresponding second segmentation region. Thus, the target segmentation result corresponding to the target point cloud data is obtained after the data of the first segmentation region and the second segmentation region are merged.

[0081] Exemplarily, feature analysis is performed on each region in the initial segmentation result to extract its corresponding target feature vector, and the extracted target feature vector is input into the trained classification model for classification prediction. If the prediction result is consistent with the type corresponding to the region in the initial segmentation result, it means that the segmentation result corresponding to the region is correct. If the prediction result is inconsistent with the type corresponding to the region in the initial segmentation result, it means that the segmentation result corresponding to the region is incorrect. It is necessary to reintegrate the region according to the first segmentation result and the second segmentation result until the target segmentation result corresponding to the region is obtained.

[0082] In some embodiments, the data classification analysis of the initial segmentation result to obtain the target segmentation result corresponding to the target point cloud data includes: determining the common point cloud data belonging to the first segmentation result and the second segmentation result, the first point cloud data belonging to the first segmentation result but not corresponding to the second segmentation result, and the second point cloud data belonging to the second segmentation result but not corresponding to the first segmentation result in each sub-segmentation result of the initial segmentation result; using the multi-scale feature extraction layer of the target classification model to perform local feature extraction on the common point cloud data to obtain first feature information; using the global feature extraction layer of the target classification model to extract the common point cloud data; The first point cloud data is subjected to global feature extraction to obtain second feature information; the first feature information and the second feature information are subjected to feature combination using the feature concatenation layer of the target classification model to obtain first combined feature information corresponding to the common point cloud data; the first combined feature information is subjected to target classification using the data classification layer of the target classification model to obtain first distribution information; the common point cloud data and the first point cloud data are subjected to local feature extraction using the multi-scale feature extraction layer of the target classification model to obtain third feature information; the common point cloud data and the first point cloud data are subjected to global feature extraction using the global feature extraction layer of the target classification model to obtain fourth feature information. The method comprises the following steps: performing feature combination on the third feature information and the fourth feature information using the feature splicing layer of the target classification model to obtain second combined feature information corresponding to the common point cloud data and the first point cloud data; performing target classification on the second combined feature information using the data classification layer of the target classification model to obtain second distribution information; performing local feature extraction on the common point cloud data and the second point cloud data using the multi-scale feature extraction layer of the target classification model to obtain fifth feature information; performing global feature extraction on the common point cloud data and the second point cloud data using the global feature extraction layer of the target classification model to obtain sixth feature information. ; Using the feature concatenation layer of the target classification model to perform feature combination on the fifth feature information and the sixth feature information to obtain the third combined feature information corresponding to the common point cloud data and the second point cloud data; Using the data classification layer of the target classification model to perform target classification on the third combined feature information to obtain third distribution information; Calculate the first distribution difference between the first distribution information and the second distribution information and calculate the second distribution difference between the first distribution information and the third distribution information; According to the first distribution difference and the second distribution difference, perform data adjustment on the initial segmentation result to obtain the target segmentation result corresponding to the target point cloud data.

[0083] Exemplarily, in each sub-segmentation result of the initial segmentation result, common point cloud data belonging to both the first segmentation result and the second segmentation result are found. In each sub-segmentation result of the initial segmentation result, first point cloud data belonging to the first segmentation result but not belonging to the second segmentation result are found. And in each sub-segmentation result of the initial segmentation result, second point cloud data belonging to the second segmentation result but not belonging to the first segmentation result are found.

[0084] Exemplarily, the multi-scale feature extraction layer of the target classification model is used to extract features from different scales to reflect the local structure and details of the common point cloud data, thereby performing local feature extraction on the common point cloud data to obtain the first feature information. The global feature extraction layer of the target classification model is used to perform global feature extraction on the common point cloud data, such as overall shape, color distribution, etc., to obtain the second feature information.

[0085] Exemplarily, the first feature information and the second feature information are feature combined using the feature concatenation layer of the target classification model to obtain first combined feature information corresponding to the common point cloud data. Thus, the first combined feature information is subjected to target classification using the data classification layer of the target classification model to obtain first distribution information. The first distribution information includes a category label and a probability distribution corresponding to the category label.

[0086] Exemplarily, the multi-scale feature extraction layer of the target classification model is used to merge the common point cloud data and the first point cloud data and perform local feature extraction to obtain the third feature information. The global feature extraction layer of the target classification model is then used to perform global feature extraction on the common point cloud data and the first point cloud data to obtain the fourth feature information. Thus, the feature concatenation layer of the target classification model is used to perform feature combination on the third feature information and the fourth feature information to obtain the second combined feature information corresponding to the common point cloud data and the first point cloud data. Finally, the data classification layer of the target classification model is used to perform target classification on the second combined feature information to obtain the second distribution information. The second distribution information includes the category label and the probability distribution corresponding to the category label.

[0087] Exemplarily, the multi-scale feature extraction layer of the target classification model is used to merge the common point cloud data and the second point cloud data and then perform local feature extraction to obtain the fifth feature information. The global feature extraction layer of the target classification model is used to perform global feature extraction on the common point cloud data and the second point cloud data to obtain the sixth feature information. The feature concatenation layer of the target classification model is then used to perform feature combination on the fifth feature information and the sixth feature information to obtain the third combined feature information corresponding to the common point cloud data and the second point cloud data. Finally, the data classification layer of the target classification model is used to perform target classification on the third combined feature information to obtain the third distribution information. The third distribution information includes the category label and the probability distribution corresponding to the category label.

[0088] Exemplarily, the difference between the first distribution information and the second distribution information is calculated using KL divergence to obtain the first distribution difference. And the difference between the first distribution information and the third distribution information is calculated using KL divergence to obtain the second distribution difference. Thus, the influence of the first point cloud data and the second point cloud data on the common point cloud data is obtained according to the first distribution difference and the second distribution difference, thereby obtaining the point cloud data corresponding to the maximum first distribution difference and the second distribution difference, that is, the point cloud data is any one of the first point cloud data and the second point cloud data, and then the point cloud data is combined with the common point cloud data to obtain the target segmentation result corresponding to the target point cloud data.

[0089] Step S106: determining a three-dimensional reconstruction result corresponding to the target case event scene according to the target segmentation result.

[0090] For example, according to the characteristics and requirements of the target case event scene, a suitable 3D reconstruction model is selected, such as a surface reconstruction, volume reconstruction, etc. 3D reconstruction model is used to determine the 3D reconstruction result corresponding to the target case event scene using the target segmentation result.

[0091] In some embodiments, determining the three-dimensional reconstruction result corresponding to the target case event scene based on the target segmentation result includes: extracting the key reference object corresponding to the target case event scene from the target segmentation result through a global optimization algorithm; extracting the target boundary points of the target object from the target segmentation result, and using a fitting algorithm to fit the target boundary points to generate a corresponding fitting curve; calculating the spatial distribution information between the target segmentation result and the key reference object; rendering and reconstructing the target case event scene according to the spatial distribution information and the fitting curve to obtain the three-dimensional reconstruction result corresponding to the target case event scene.

[0092] Exemplarily, a global optimization algorithm such as Monte Carlo method or particle swarm optimization is used to identify key reference objects corresponding to the target case event scene from the target segmentation results.

[0093] Exemplarily, the target boundary points of the target object are extracted from the target segmentation result by an edge detection algorithm. The target boundary points are usually the intersection points between the target object and other areas or backgrounds. Therefore, a fitting algorithm such as Bezier curve fitting is used to fit the extracted target boundary points, and then the parameters of the fitting algorithm are adjusted, such as the number of control points, smoothness, fitting error, etc., to obtain the best fitting curve, and then the fitting curve of the target object is generated according to the fitting algorithm and the adjusted parameters. Ensure that the fitting curve can accurately describe the boundary shape of the target object.

[0094] Exemplarily, the spatial distribution information is defined, which usually includes the position, direction, shape and other information of the target object in the target segmentation result. The spatial distribution information can be represented by the statistical characteristics of the target point cloud data (such as mean, covariance matrix). Thus, the spatial distribution information between the target segmentation result and the key reference object is calculated. The distance, angle, overlapping area and other information between the target objects can be included, and the calculated spatial distribution information is then represented in numerical or matrix form for subsequent rendering and reconstruction.

[0095] For example, a rendering algorithm such as rasterization rendering is selected to render and reconstruct the target case event scene, apply the calculated spatial distribution information to the rendering process, adjust the rendering position, direction, shape, etc. of the object, and ensure that the rendering result is consistent with the actual scene. Thus, according to the rendering algorithm and the adjusted parameters, a three-dimensional reconstruction result of the target case event scene is generated. Ensure that the rendering result can accurately reflect the target segmentation result and the spatial distribution information of the key reference objects.

[0096] In some embodiments, after obtaining the three-dimensional reconstruction result, the method further includes: performing spatial measurement of the target case event scene according to the three-dimensional reconstruction result to obtain a measurement result; performing physical evidence analysis on the target case event scene according to the three-dimensional reconstruction result to obtain an analysis result; performing trajectory deduction on the target case event scene according to the three-dimensional reconstruction result to obtain a deduction result; and determining a target analysis result corresponding to the target case event scene according to the measurement result, the analysis result, and the deduction result.

[0097] Exemplarily, a three-dimensional distance measurement tool is used to measure the straight-line distance between objects, a three-dimensional angle measurement tool is used to measure the angle between objects, and a three-dimensional volume measurement tool is used to measure the volume of objects, thereby determining the target to be measured, such as the distance between objects, the size of objects, the volume of objects, etc., and then obtaining measurement results, which may include information such as distance, angle, and volume.

[0098] Exemplarily, features of physical evidence are extracted from the three-dimensional reconstruction results, such as shape, size, color, texture, etc. The features of physical evidence can be extracted using image processing technology or a three-dimensional feature extraction algorithm, or the extracted physical evidence features can be compared with a known physical evidence database to determine the source and attributes of the physical evidence, thereby obtaining analysis results.

[0099] Exemplarily, trajectory data, such as the initial and final positions of physical evidence, the movement paths of personnel, etc., are extracted from the three-dimensional reconstruction results, and then trajectory deduction is performed based on the trajectory model and the extracted trajectory data. Trajectory deduction can be performed through numerical simulation, physical simulation or logical reasoning methods to obtain deduction results. Finally, the measurement results, analysis results and deduction results are integrated to obtain the corresponding target analysis results. Thereby, spatial measurement, physical evidence analysis and trajectory deduction of the target case event scene are achieved based on the three-dimensional reconstruction results, and the target analysis results are finally determined, so that they can be used for case event investigation.

[0100] See also Figure 2 , Figure 2 The present application provides a 3D laser-based crime scene 3D reconstruction system 200, which includes a data acquisition module 201, a data filtering module 202, a first segmentation module 203, a second segmentation module 204, a data fusion module 205, and a 3D reconstruction module 206. The data acquisition module 201 is used to determine the target crime scene and use the target sensor to perform data acquisition on the target crime scene to obtain initial point cloud data; the data filtering module 202 is used to perform data filtering on the initial point cloud data. The target point cloud data is obtained by wave processing; the first segmentation module 203 is used to perform data segmentation on the target point cloud data using a density-based clustering algorithm to obtain a first segmentation result; the second segmentation module 204 is used to perform data segmentation on the target point cloud data using a growth direction and color fusion strategy to obtain a second segmentation result; the data fusion module 205 is used to fuse the first segmentation result and the second segmentation result to obtain a target segmentation result corresponding to the target point cloud data; the three-dimensional reconstruction module 206 is used to determine the three-dimensional reconstruction result corresponding to the target case event scene according to the target segmentation result.

[0101] In some implementations, the data filtering module 202 performs, during the process of performing data filtering on the initial point cloud data to obtain target point cloud data:

[0102] Performing a neighbor search on the initial point cloud data to obtain first associated neighbor data corresponding to each first sub-point cloud data in the initial point cloud data;

[0103] Calculating a first distance between the first sub-point cloud data and the first associated neighbor data, and determining a second distance and a standard deviation value corresponding to the initial point cloud data according to the first distance;

[0104] Performing noise removal on the initial point cloud data according to the second distance and the standard deviation value to obtain denoised point cloud data;

[0105] Performing a neighbor search on the denoised point cloud data to obtain second associated neighbor data corresponding to each second sub-point cloud data in the denoised point cloud data;

[0106] Performing center calculation according to the second associated neighbor data to obtain a target center corresponding to the second sub-point cloud data;

[0107] Performing linear fitting based on the second associated neighbor data and the target center to obtain target straight line parameters corresponding to the second sub-point cloud data, wherein the target straight line parameters include a slope coefficient and a constant coefficient;

[0108] Performing data filtering on the second sub-point cloud data according to the target straight line parameter to obtain the target point cloud data;

[0109] The slope coefficient and the constant coefficient are obtained according to the following formula:

[0110]

[0111]

[0112] in, represents the slope coefficient corresponding to the i-th second sub-point cloud data, represents the constant coefficient corresponding to the i-th second sub-point cloud data, n represents the number of neighbors corresponding to the second associated neighbor data, represents the distance between the i-th second sub-point cloud data and the j-th second associated neighbor data, represents the target center corresponding to the i-th second sub-point cloud data, represents the smoothing parameter.

[0113] In some implementations, the first segmentation module 203 performs, during the process of segmenting the target point cloud data using the density-based clustering algorithm to obtain the first segmentation result:

[0114] Performing a neighbor search on the target point cloud data to obtain third associated neighbor data corresponding to each third sub-point cloud data in the target point cloud data;

[0115] Calculating a third distance between the third sub-point cloud data and each first sub-neighborhood data in the third associated neighbor data;

[0116] Determine a neighbor radius corresponding to the third sub-point cloud data according to the third distance, and perform data clustering on the target point cloud data according to the neighbor radius and the third sub-point cloud data to obtain an initial clustering result;

[0117] Performing a mutually reachable density test on the fourth sub-point cloud data of each sub-cluster in the initial clustering result to obtain a target clustering result;

[0118] The target point cloud data is segmented according to the target clustering result to obtain the first segmentation result.

[0119] In some implementations, during the process of determining the neighbor radius corresponding to the third sub-point cloud data according to the third distance, the first segmentation module 203 executes:

[0120] Calculate a first distance difference between adjacent first sub-neighborhood data in the third associated neighborhood data according to the third distance;

[0121] Performing data screening on the third associated neighbor data according to the first distance difference and a preset threshold to obtain fourth associated neighbor data;

[0122] Obtaining a fourth distance between the fourth associated neighbor data and the third sub-point cloud data from the third distance;

[0123] Calculate the second distance difference between adjacent second sub-neighborhood data in the fourth associated neighborhood data according to the fourth distance;

[0124] Determine the neighbor radius corresponding to the third sub-point cloud data according to the second distance difference and the fourth distance;

[0125] The nearest neighbor radius is obtained according to the following formula:

[0126]

[0127] in, represents the neighbor radius corresponding to the i-th third sub-point cloud data, s represents the number of neighbors corresponding to the fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the j-th fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the j-1-th fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the first fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the s-th fourth associated neighbor data, Represents the second distance difference between adjacent second sub-neighborhood data.

[0128] In some implementations, the second segmentation module 204, in the process of segmenting the target point cloud data using the growth direction and color fusion strategy to obtain the second segmentation result, executes:

[0129] voxelize the target point cloud data to obtain supervoxel data, and determine a seed voxel and adjacent voxels corresponding to the seed voxel according to the supervoxel data;

[0130] Calculating the normal angle and color difference between the seed voxel and the adjacent voxel;

[0131] Merging data of the seed voxel and the adjacent voxel according to the normal angle and the color difference to obtain a regional growth direction corresponding to the seed voxel;

[0132] Performing regional segmentation on the supervoxel data according to the regional growth direction and the seed voxel to obtain a voxel segmentation result;

[0133] The second segmentation result corresponding to the target point cloud data is determined according to the voxel segmentation result.

[0134] In some implementations, the data fusion module 205 performs, during the process of fusing the first segmentation result and the second segmentation result to obtain the target segmentation result corresponding to the target point cloud data:

[0135] Merging the first segmentation result and the second segmentation result to obtain an initial segmentation result;

[0136] Perform data classification analysis on the initial segmentation result to obtain the target segmentation result corresponding to the target point cloud data.

[0137] In some implementations, the data fusion module 205 performs, during the process of performing data classification analysis on the initial segmentation result to obtain the target segmentation result corresponding to the target point cloud data:

[0138] Determine, in each sub-segmentation result of the initial segmentation result, common point cloud data belonging to the first segmentation result and the second segmentation result, first point cloud data belonging to the first segmentation result but not corresponding to the second segmentation result, and second point cloud data belonging to the second segmentation result but not corresponding to the first segmentation result;

[0139] Using a multi-scale feature extraction layer of a target classification model to perform local feature extraction on the common point cloud data to obtain first feature information;

[0140] Performing global feature extraction on the common point cloud data using the global feature extraction layer of the target classification model to obtain second feature information;

[0141] Using the feature concatenation layer of the target classification model, the first feature information and the second feature information are feature combined to obtain first combined feature information corresponding to the common point cloud data;

[0142] Using the data classification layer of the target classification model to perform target classification on the first combined feature information to obtain first distribution information;

[0143] Using the multi-scale feature extraction layer of the target classification model to perform local feature extraction on the common point cloud data and the first point cloud data to obtain third feature information;

[0144] Using the global feature extraction layer of the target classification model to perform global feature extraction on the common point cloud data and the first point cloud data to obtain fourth feature information;

[0145] Using the feature concatenation layer of the target classification model, the third feature information and the fourth feature information are feature combined to obtain second combined feature information corresponding to the common point cloud data and the first point cloud data;

[0146] Using the data classification layer of the target classification model to perform target classification on the second combined feature information to obtain second distribution information;

[0147] Using the multi-scale feature extraction layer of the target classification model to perform local feature extraction on the common point cloud data and the second point cloud data to obtain fifth feature information;

[0148] Performing global feature extraction on the common point cloud data and the second point cloud data using the global feature extraction layer of the target classification model to obtain sixth feature information;

[0149] Using the feature concatenation layer of the target classification model, the fifth feature information and the sixth feature information are feature combined to obtain third combined feature information corresponding to the common point cloud data and the second point cloud data;

[0150] Using the data classification layer of the target classification model to perform target classification on the third combined feature information to obtain third distribution information;

[0151] calculating a first distribution difference between the first distribution information and the second distribution information and calculating a second distribution difference between the first distribution information and the third distribution information;

[0152] The initial segmentation result is adjusted according to the first distribution difference and the second distribution difference to obtain the target segmentation result corresponding to the target point cloud data.

[0153] In some implementations, the 3D reconstruction module 206, during the process of determining the 3D reconstruction result corresponding to the target case event scene according to the target segmentation result, performs:

[0154] The key reference object corresponding to the target case event scene is obtained from the target segmentation result by a global optimization algorithm;

[0155] Extracting target boundary points of the target object from the target segmentation result, and fitting the target boundary points using a fitting algorithm to generate a corresponding fitting curve;

[0156] Calculating spatial distribution information between the target segmentation result and the key reference object;

[0157] The target case event scene is rendered and reconstructed according to the spatial distribution information and the fitting curve to obtain the three-dimensional reconstruction result corresponding to the target case event scene.

[0158] In some implementations, after obtaining the 3D reconstruction result, the 3D reconstruction module 206 further executes:

[0159] Performing spatial measurement on the target case event scene according to the three-dimensional reconstruction result to obtain a measurement result;

[0160] Performing physical evidence analysis on the target case incident scene according to the three-dimensional reconstruction result to obtain an analysis result;

[0161] Performing trajectory deduction on the target case event scene according to the three-dimensional reconstruction result to obtain a deduction result;

[0162] A target analysis result corresponding to the target case event scene is determined according to the measurement result, the analysis result and the deduction result.

[0163] In some embodiments, the three-dimensional laser-based incident scene three-dimensional reconstruction system 200 can be applied to a terminal device.

[0164] It should be noted that, those skilled in the art can clearly understand that, for the sake of convenience and brevity of description, the specific working process of the three-dimensional laser-based crime scene three-dimensional reconstruction system 200 described above can refer to the corresponding process in the aforementioned three-dimensional laser-based crime scene three-dimensional reconstruction method embodiment, and will not be repeated here.

[0165] An embodiment of the present invention also provides a storage medium for computer-readable storage, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any one of the three-dimensional laser-based incident scene three-dimensional reconstruction methods provided in the description of the embodiment of the present invention.

[0166] The storage medium may be an internal storage unit of the terminal device described in the foregoing embodiment, such as a hard disk or memory of the terminal device. The storage medium may also be an external storage device of the terminal device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc., equipped on the terminal device.

[0167] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In a hardware embodiment, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or transient medium). As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0168] It should be understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0169] The serial numbers of the embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A three-dimensional reconstruction method of a crime scene based on three-dimensional laser, characterized in that: The method comprises: Determine the target case event scene, and use the target sensor to collect data on the target case event scene to obtain initial point cloud data; Performing data filtering processing on the initial point cloud data to obtain target point cloud data; Using a density-based clustering algorithm to perform data segmentation on the target point cloud data to obtain a first segmentation result; Performing data segmentation on the target point cloud data using a fusion strategy of growth direction and color to obtain a second segmentation result; fusing the first segmentation result and the second segmentation result to obtain a target segmentation result corresponding to the target point cloud data; Determine a three-dimensional reconstruction result corresponding to the target case event scene according to the target segmentation result; The step of performing data filtering on the initial point cloud data to obtain target point cloud data includes: Performing a neighbor search on the initial point cloud data to obtain first associated neighbor data corresponding to each first sub-point cloud data in the initial point cloud data; Calculating a first distance between the first sub-point cloud data and the first associated neighbor data, and determining a second distance and a standard deviation value corresponding to the initial point cloud data according to the first distance; Performing noise removal on the initial point cloud data according to the second distance and the standard deviation value to obtain denoised point cloud data; Performing a neighbor search on the denoised point cloud data to obtain second associated neighbor data corresponding to each second sub-point cloud data in the denoised point cloud data; Performing center calculation according to the second associated neighbor data to obtain a target center corresponding to the second sub-point cloud data; Performing linear fitting based on the second associated neighbor data and the target center to obtain target straight line parameters corresponding to the second sub-point cloud data, wherein the target straight line parameters include a slope coefficient and a constant coefficient; Performing data filtering on the second sub-point cloud data according to the target straight line parameter to obtain the target point cloud data; The slope coefficient and the constant coefficient are obtained according to the following formula: ; ; in, represents the slope coefficient corresponding to the i-th second sub-point cloud data, represents the constant coefficient corresponding to the i-th second sub-point cloud data, n represents the number of neighbors corresponding to the second associated neighbor data, represents the distance between the i-th second sub-point cloud data and the j-th second associated neighbor data, represents the target center corresponding to the i-th second sub-point cloud data, represents the smoothing parameter.

2. The method according to claim 1, characterized in that The step of performing data segmentation on the target point cloud data using a density-based clustering algorithm to obtain a first segmentation result includes: Performing a neighbor search on the target point cloud data to obtain third associated neighbor data corresponding to each third sub-point cloud data in the target point cloud data; Calculating a third distance between the third sub-point cloud data and each first sub-neighborhood data in the third associated neighbor data; Determine a neighbor radius corresponding to the third sub-point cloud data according to the third distance, and perform data clustering on the target point cloud data according to the neighbor radius and the third sub-point cloud data to obtain an initial clustering result; Performing a mutually reachable density test on the fourth sub-point cloud data of each sub-cluster in the initial clustering result to obtain a target clustering result; The target point cloud data is segmented according to the target clustering result to obtain the first segmentation result.

3. The method according to claim 2, characterized in that The determining, according to the third distance, a neighbor radius corresponding to the third sub-point cloud data includes: Calculate a first distance difference between adjacent first sub-neighborhood data in the third associated neighborhood data according to the third distance; Performing data screening on the third associated neighbor data according to the first distance difference and a preset threshold to obtain fourth associated neighbor data; Obtaining a fourth distance between the fourth associated neighbor data and the third sub-point cloud data from the third distance; Calculate the second distance difference between adjacent second sub-neighborhood data in the fourth associated neighborhood data according to the fourth distance; Determine the neighbor radius corresponding to the third sub-point cloud data according to the second distance difference and the fourth distance; The nearest neighbor radius is obtained according to the following formula: ; in, represents the neighbor radius corresponding to the i-th third sub-point cloud data, s represents the number of neighbors corresponding to the fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the j-th fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the j-1-th fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the first fourth associated neighbor data, represents the fourth distance between the i-th third sub-point cloud data and the s-th fourth associated neighbor data, Represents the second distance difference between adjacent second sub-neighborhood data.

4. The method according to claim 1, characterized in that The step of segmenting the target point cloud data using a fusion strategy of growth direction and color to obtain a second segmentation result includes: voxelize the target point cloud data to obtain supervoxel data, and determine a seed voxel and adjacent voxels corresponding to the seed voxel according to the supervoxel data; Calculating the normal angle and color difference between the seed voxel and the adjacent voxel; Merging data of the seed voxel and the adjacent voxel according to the normal angle and the color difference to obtain a regional growth direction corresponding to the seed voxel; Performing regional segmentation on the supervoxel data according to the regional growth direction and the seed voxel to obtain a voxel segmentation result; The second segmentation result corresponding to the target point cloud data is determined according to the voxel segmentation result.

5. The method according to claim 1, characterized in that The fusing the first segmentation result and the second segmentation result to obtain the target segmentation result corresponding to the target point cloud data includes: Merging the first segmentation result and the second segmentation result to obtain an initial segmentation result; Perform data classification analysis on the initial segmentation result to obtain the target segmentation result corresponding to the target point cloud data.

6. The method according to claim 5, characterized in that The performing data classification analysis on the initial segmentation result to obtain the target segmentation result corresponding to the target point cloud data includes: Determine, in each sub-segmentation result of the initial segmentation result, common point cloud data belonging to the first segmentation result and the second segmentation result, first point cloud data belonging to the first segmentation result but not corresponding to the second segmentation result, and second point cloud data belonging to the second segmentation result but not corresponding to the first segmentation result; Using a multi-scale feature extraction layer of a target classification model to perform local feature extraction on the common point cloud data to obtain first feature information; Performing global feature extraction on the common point cloud data using the global feature extraction layer of the target classification model to obtain second feature information; Using the feature concatenation layer of the target classification model, the first feature information and the second feature information are feature combined to obtain first combined feature information corresponding to the common point cloud data; Using the data classification layer of the target classification model to perform target classification on the first combined feature information to obtain first distribution information; Using the multi-scale feature extraction layer of the target classification model to perform local feature extraction on the common point cloud data and the first point cloud data to obtain third feature information; Using the global feature extraction layer of the target classification model to perform global feature extraction on the common point cloud data and the first point cloud data to obtain fourth feature information; Using the feature concatenation layer of the target classification model, the third feature information and the fourth feature information are feature combined to obtain second combined feature information corresponding to the common point cloud data and the first point cloud data; Using the data classification layer of the target classification model to perform target classification on the second combined feature information to obtain second distribution information; Using the multi-scale feature extraction layer of the target classification model to perform local feature extraction on the common point cloud data and the second point cloud data to obtain fifth feature information; Performing global feature extraction on the common point cloud data and the second point cloud data using the global feature extraction layer of the target classification model to obtain sixth feature information; Using the feature concatenation layer of the target classification model, the fifth feature information and the sixth feature information are feature combined to obtain third combined feature information corresponding to the common point cloud data and the second point cloud data; Using the data classification layer of the target classification model to perform target classification on the third combined feature information to obtain third distribution information; calculating a first distribution difference between the first distribution information and the second distribution information and calculating a second distribution difference between the first distribution information and the third distribution information; The initial segmentation result is adjusted according to the first distribution difference and the second distribution difference to obtain the target segmentation result corresponding to the target point cloud data.

7. The method according to claim 1, characterized in that Determining the three-dimensional reconstruction result corresponding to the target case event scene according to the target segmentation result includes: The key reference object corresponding to the target case event scene is obtained from the target segmentation result by a global optimization algorithm; Extracting target boundary points of the target object from the target segmentation result, and fitting the target boundary points using a fitting algorithm to generate a corresponding fitting curve; Calculating spatial distribution information between the target segmentation result and the key reference object; The target case event scene is rendered and reconstructed according to the spatial distribution information and the fitting curve to obtain the three-dimensional reconstruction result corresponding to the target case event scene.

8. The method according to any one of claims 1 to 7, characterized in that After obtaining the three-dimensional reconstruction result, the method further includes: Performing spatial measurement on the target case event scene according to the three-dimensional reconstruction result to obtain a measurement result; Performing physical evidence analysis on the target case incident scene according to the three-dimensional reconstruction result to obtain an analysis result; Performing trajectory deduction on the target case event scene according to the three-dimensional reconstruction result to obtain a deduction result; A target analysis result corresponding to the target case event scene is determined according to the measurement result, the analysis result and the deduction result.

9. A three-dimensional reconstruction system for crime scene based on three-dimensional laser, characterized in that: include: A data acquisition module is used to determine the target case event scene and use a target sensor to collect data from the target case event scene to obtain initial point cloud data; A data filtering module is used to perform data filtering processing on the initial point cloud data to obtain target point cloud data; wherein, the data filtering processing on the initial point cloud data to obtain target point cloud data includes: performing a neighbor search on the initial point cloud data to obtain first associated neighbor data corresponding to each first sub-point cloud data in the initial point cloud data; calculating a first distance between the first sub-point cloud data and the first associated neighbor data, and determining a second distance and a standard deviation value corresponding to the initial point cloud data according to the first distance; performing noise removal on the initial point cloud data according to the second distance and the standard deviation value to obtain a denoised point cloud data; performing a neighbor search on the denoised point cloud data to obtain second associated neighbor data corresponding to each second sub-point cloud data in the denoised point cloud data; performing center calculation according to the second associated neighbor data to obtain a target center corresponding to the second sub-point cloud data; performing linear fitting according to the second associated neighbor data and the target center to obtain target straight line parameters corresponding to the second sub-point cloud data, the target straight line parameters including a slope coefficient and a constant coefficient; performing data filtering on the second sub-point cloud data according to the target straight line parameters to obtain the target point cloud data; wherein the slope coefficient and the constant coefficient are obtained according to the following formula: ; ; in, represents the slope coefficient corresponding to the i-th second sub-point cloud data, represents the constant coefficient corresponding to the i-th second sub-point cloud data, n represents the number of neighbors corresponding to the second associated neighbor data, represents the distance between the i-th second sub-point cloud data and the j-th second associated neighbor data, represents the target center corresponding to the i-th second sub-point cloud data, represents the smoothing parameter; A first segmentation module, used to perform data segmentation on the target point cloud data using a density-based clustering algorithm to obtain a first segmentation result; A second segmentation module, used for performing data segmentation on the target point cloud data by using a fusion strategy of growth direction and color to obtain a second segmentation result; A data fusion module, used for fusing the first segmentation result and the second segmentation result to obtain a target segmentation result corresponding to the target point cloud data; The three-dimensional reconstruction module is used to determine the three-dimensional reconstruction result corresponding to the target case event scene according to the target segmentation result.

Images