CN-121982483-A - Legal medical expert injury analysis method and system based on three-dimensional modeling and visual language big model

CN121982483ACN 121982483 ACN121982483 ACN 121982483ACN-121982483-A

Abstract

The invention provides a forensic injury analysis method and a forensic injury analysis system based on three-dimensional modeling and a visual language large model, and relates to the technical field of data processing, wherein the method comprises the steps of collecting an original three-dimensional model of a wound part by using a structured light scanner, and preprocessing the original three-dimensional model to obtain a standardized three-dimensional model of the wound; the method comprises the steps of identifying three wound edge vertexes of a wound edge in a standardized wound three-dimensional model, establishing a reference topological relation, conducting multi-level structure subdivision on the reference topological relation based on a deep feature extraction mechanism of a convolutional neural network to obtain a subdivided topological structure, conducting deep extraction and quantitative analysis on wound deformation features of the subdivided topological structure, accurately capturing feature information of micro-deformation and micro-topological structure of a wound, and obtaining a proportional calibration parameter by combining feature information obtained by extraction and analysis. The invention realizes accurate quantitative extraction of wound characteristics, standardized data processing and intelligent wound identification reasoning.

Inventors

CHENG JIANBO
YING JIANBO
DING SHAOCHENG
LIU JUN
Zhou Jiachuan
LI YANG
NIE ZHIJIAN
ZHANG GUOJIAN
Zhu Guantao

Assignees

杭州云枢智能科技有限公司
杭州市公安局

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. The forensic injury analysis method based on the three-dimensional modeling and the visual language big model is characterized by comprising the following steps of: Collecting an original three-dimensional model of a wound part by using a structured light scanner, and preprocessing the original three-dimensional model to obtain a standardized wound three-dimensional model; Identifying three wound edge vertexes of a wound edge in a standardized wound three-dimensional model, establishing a reference topological relation, and carrying out multi-level structure subdivision on the reference topological relation based on a deep feature extraction mechanism of a convolutional neural network to obtain a subdivided topological structure; Deep extraction and quantitative analysis of wound deformation characteristics are carried out on the subdivided topological structure, characteristic information of small deformation and small topological structure of the wound is accurately captured, and the proportional calibration parameters are obtained through calculation by combining the characteristic information obtained through extraction and analysis; Performing geometric pose normalization and scale standardization treatment on the wound three-dimensional model according to the proportion calibration parameters to obtain a treated wound three-dimensional model; According to the quantized and marked three-dimensional model, performing multi-view standardized shooting under a world coordinate system through a virtual camera to obtain a two-dimensional image containing wound morphology, a physical reference point and quantized and marked information; Inputting the two-dimensional image and the associated quantized labeling data into a visual language large model subjected to legal medical expert injury data fine adjustment, extracting visual features and quantized features of wounds, and completing judgment of a wound formation time interval and identification of organic injury based on the extracted features.
2. The method for forensic injury analysis based on three-dimensional modeling and visual language big model according to claim 1, wherein the steps of collecting an original three-dimensional model of a wound site by using a structured light scanner, and preprocessing the original three-dimensional model to eliminate scanning noise and repair data defects to obtain a standardized three-dimensional model of the wound, comprise: Receiving an original three-dimensional model uploaded by a structured light scanner, and separating and extracting point cloud data and an initial grid model in the original three-dimensional model; Calculating the distance mean value and standard deviation of each cloud point in the neighborhood range by performing statistical filtering processing on the point cloud data, and removing outliers with deviation mean values exceeding a preset multiple standard deviation to obtain the point cloud data after statistical filtering; performing radius filtering processing on the point cloud data, counting the number of neighborhood points of each point cloud point in a set searching radius, and removing isolated points with the number of the neighborhood points lower than a preset threshold value to obtain clean point cloud data with the denoising completed; Reconstructing a surface grid based on the pure point cloud data, performing Laplace grid smoothing on the reconstructed grid model, and eliminating burrs and saw-tooth distortion on the grid surface to obtain a smoothed grid model; And detecting a topological cavity area in the smooth grid model, and filling the cavity area with grids to obtain a complete and continuous standardized wound three-dimensional model without missing.
3. The method for forensic injury analysis based on three-dimensional modeling and visual language big model according to claim 2, wherein the steps of carrying out multi-level structure subdivision on the reference topological relation based on a deep feature extraction mechanism of a convolutional neural network by identifying three wound edge vertices of a wound edge in a standardized wound three-dimensional model and establishing the reference topological relation to obtain a subdivided topological structure comprise the following steps: According to a standardized wound three-dimensional model, based on curvature mutation characteristics of wound surfaces and surrounding normal skin and forensic anatomy priori knowledge, three wound edge vertexes with geometric characteristics are automatically positioned and identified at the edge of the wound, and the three wound edge vertexes respectively correspond to a starting endpoint, a terminating endpoint and a morphological turning key point of the wound to obtain vertex sets of three wound edge vertex space coordinates; Based on the vertex set, constructing an initial topology framework of the wound edge by taking three wound edge vertices as references, and forming a reference topology relation representing the overall outline of the wound by generating initial connecting edges among the three vertices to obtain reference topology data; Mapping the reference topological data into a two-dimensional parameterized grid, inputting the two-dimensional parameterized grid into a pre-trained convolutional neural network feature extractor, and gradually extracting local geometric features and global morphological features of a topological structure through a convolutional layer to obtain a feature map of multi-level feature response; Based on the multi-level characteristic spectrum, the geometric discontinuous region and the morphological change region of the wound edge are adaptively identified, adaptive subdivision is implemented on the reference topological relation according to the characteristic response intensity, subdivision density is increased in the region with severe curvature change and the morphological complex region, sparse connection is kept in the smooth region, and the refined topological structure is obtained.
4. The forensic injury analysis method based on three-dimensional modeling and visual language big model according to claim 3 is characterized in that the deep extraction and quantitative analysis of wound deformation characteristics are carried out on the subdivided topological structure, the characteristic information of the tiny deformation and the tiny topological structure of the wound is accurately captured, and the proportional calibration parameters are obtained by combining the characteristic information obtained by extraction and analysis, and the method comprises the following steps: according to the refined topological structure, calculating local geometric deformation characteristics point by point along a subdivision topological path of the wound edge, wherein the local geometric deformation characteristics comprise main curvature, normal vector deflection angles and Gaussian curvature change gradients at each subdivision node to obtain a characteristic vector sequence; Identifying deformation continuity features and discontinuous mutation points of the wound edge under a microscopic scale by carrying out time sequence quantitative analysis on the feature vector sequence, screening out an effective deformation feature subset representing the real edge morphology of the wound based on morphological priori knowledge, and eliminating pseudo deformation interference caused by scanning residual noise to obtain an effective deformation feature set; based on the effective deformation characteristic set, a mapping relation between the deformation characteristics of the wound edge and a standard reference scale is constructed, and a proportional calibration parameter for compensating scale distortion caused by non-rigid deformation of the wound is calculated by analyzing a relative deformation distribution rule of the wound edge in a reference frame formed by connecting three wound edge vertexes, wherein the proportional calibration parameter comprises a linear scaling factor along the long axis direction of the wound and a non-uniform deformation compensation coefficient along the normal direction of the wound edge.
5. The method for forensic injury analysis based on three-dimensional modeling and visual language big model according to claim 4 is characterized in that geometric pose normalization and scale standardization processing are carried out on a wound three-dimensional model according to proportion calibration parameters to obtain a processed wound three-dimensional model, quantitative measurement and labeling of wounds are carried out on the processed wound three-dimensional model, length and area data of the wounds are obtained, and the quantized and labeled three-dimensional model is obtained, and the method comprises the following steps: Performing scale correction on the model in the long axis direction of the wound through a linear scaling factor, compensating local deformation of the model along the normal direction of the wound edge through a non-uniform deformation compensation coefficient so as to eliminate scale distortion caused by non-rigid deformation, and adjusting the model to a standard anatomical posture to obtain a wound three-dimensional model subjected to geometric posture normalization and scale normalization treatment; The method comprises the steps of automatically executing quantitative measurement on a processed wound three-dimensional model, wherein the measurement comprises the steps of calculating Euclidean distance between two end points of the longest axis of a wound surface as wound length, calculating the surface area of a wound surface grid as wound surface area, and simultaneously, optionally calculating the maximum depth of the wound and the average wound edge width as supplementary quantitative data so as to obtain a quantitative measurement result; Marking the quantized measurement results on the three-dimensional model in a unified visual mode, wherein the marking content at least comprises a length value, a measurement path, an area value and a boundary range, and the marked model and the quantized data are stored in a correlated mode to form the quantized and marked three-dimensional model.
6. The forensic injury analysis method based on the three-dimensional modeling and visual language big model according to claim 5 is characterized in that according to the quantized and marked three-dimensional model, multi-view standardized shooting is carried out under a world coordinate system through a virtual camera to obtain a two-dimensional image containing wound morphology, physical reference points and quantized and marked information, and the method comprises the following steps: Registering the quantized and marked three-dimensional model to a unified world coordinate system to obtain a registered three-dimensional model, taking the geometric center of a wound surface in the registered three-dimensional model as a coordinate origin, and establishing a space coordinate axis according to a main plane of the wound surface and a depth direction to establish a standardized space reference; According to the three-dimensional space azimuth of the wound surface in the standardized space reference, planning and generating a group of standardized shooting visual angles covering key anatomical structures and corresponding camera poses; Based on the planned camera pose, driving a virtual camera to perform multi-view rendering shooting on the registered three-dimensional model, and jointly projecting the model geometry and associated three-dimensional visual annotation information to a two-dimensional imaging plane to generate a group of original view images containing wound morphology, physical reference points and quantitative annotation information; and carrying out standardized post-processing on the original visual angle image, uniformly adjusting the brightness and contrast of the original visual angle image, embedding drawing information containing case identification, shooting visual angle and key quantized data, and finally outputting a group of two-dimensional image sequences meeting preset imaging standards.
7. The forensic injury analysis method based on three-dimensional modeling and a visual language big model according to claim 6, wherein inputting the two-dimensional image and associated quantitative labeling data into the visual language big model subjected to forensic injury data fine adjustment, extracting visual features and quantitative features of wounds, and completing the determination of a time interval for forming wounds and the identification of organic injuries based on the extracted features comprises: The standardized two-dimensional image sequence and the quantitative labeling data associated with the sequence are used as multi-mode input and loaded into a visual language large model which is subjected to fine adjustment by using medical injury data of a legal medical expert; Based on the loaded multi-mode input, the visual language big model extracts visual features in the two-dimensional image and quantization features in the quantization labeling data in parallel, and fuses the visual features and the quantization features to obtain comprehensive feature vectors containing wound colors, morphologies, tissue states and geometric dimensions; Calculating and judging an interval to which the wound forming time belongs according to the internal pre-learned wound healing characteristics and the time mapping relation by the comprehensive characteristic vector and the visual language big model to obtain the confidence coefficient of a judging result; Based on the time interval judgment, the visual language large model further completes the identification and classification of the organic injury according to the information related to the tissue depth and the injury range in the comprehensive feature vector, and finally outputs a structured identification report containing the time interval judgment, the organic injury identification conclusion and the corresponding confidence coefficient.
8. Forensic injury analysis system based on three-dimensional modeling and visual language big model, which implements the method according to any one of claims 1 to 7, characterized by comprising: The acquisition module is used for acquiring an original three-dimensional model of the wound part by using the structured light scanner and preprocessing the original three-dimensional model to obtain a standardized three-dimensional model of the wound; The division module is used for carrying out multi-level structure subdivision on the reference topological relation based on a deep feature extraction mechanism of the convolutional neural network by identifying three wound edge vertexes of a wound edge in the standardized wound three-dimensional model and establishing the reference topological relation to obtain a subdivided topological structure; The extraction module is used for accurately capturing the characteristic information of the tiny deformation and the tiny topological structure of the wound through deep extraction and quantitative analysis of the wound deformation characteristics of the subdivided topological structure, and obtaining the proportional calibration parameters by combining the characteristic information obtained by extraction and analysis; the quantitative module is used for carrying out geometric pose normalization and scale standardization treatment on the wound three-dimensional model according to the proportion calibration parameters to obtain a treated wound three-dimensional model; the shooting module is used for carrying out multi-view standardized shooting under a world coordinate system through a virtual camera according to the quantized and marked three-dimensional model to obtain a two-dimensional image containing wound morphology, a physical reference point and quantized and marked information; The processing module is used for inputting the two-dimensional image and the associated quantized labeling data into a visual language large model finely tuned by forensic injury data, extracting visual features and quantized features of wounds, and completing judgment of a time interval for forming the wounds and identification of organic injuries based on the extracted features.
9. A computing device, comprising: One or more processors; Storage means for storing one or more programs which when executed by the one or more processors cause the one or more processors to implement the method of any of claims 1to 7.
10. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a program which, when executed by a processor, implements the method according to any of claims 1 to 7.

Description

Legal medical expert injury analysis method and system based on three-dimensional modeling and visual language big model Technical Field The invention relates to the technical field of data processing, in particular to a forensic injury analysis method and system based on three-dimensional modeling and visual language big models. Background In the traditional forensic injury identification scene, due to the lack of accurate three-dimensional data support and intelligent analysis tools, the identification result often depends on visual observation, manual measurement and subjective experience judgment of an identification person, and has technical limitations. For example, in the conventional identification process, an identifier observes the shape of a wound through a two-dimensional photo taken on site, manually measures the length of the wound by using a ruler, determines the time for forming the wound in combination with clinical medical records and personal experience, and does not find out organic damage, but confirms the wound through operation, monitors and records the time for restoring the wound, so that serious flaws appear in the initial identification conclusion due to factor deviation and experience misjudgment, the three-dimensional topological structure and depth characteristics of the wound cannot be completely presented by a two-dimensional image, the manual measurement is easily influenced by angles and operation methods, the precision is insufficient, key information such as tiny deformation and tiny tissue damage of the wound are easily ignored, and meanwhile, the determination of the time for forming the wound and the organic damage lacks a standardized characteristic extraction and quantitative analysis mechanism, so that the accuracy of an identification result is difficult to ensure. Disclosure of Invention The technical problem to be solved by the invention is to provide a method and a system for analyzing forensic traumatology based on three-dimensional modeling and visual language big model, which realize accurate quantitative extraction of wound characteristics, standardized data processing and intelligent injury discrimination reasoning. In order to solve the technical problems, the technical scheme of the invention is as follows: in a first aspect, a forensic injury analysis method based on three-dimensional modeling and a visual language big model, the method comprising: Collecting an original three-dimensional model of a wound part by using a structured light scanner, and preprocessing the original three-dimensional model to obtain a standardized wound three-dimensional model; Identifying three wound edge vertexes of a wound edge in a standardized wound three-dimensional model, establishing a reference topological relation, and carrying out multi-level structure subdivision on the reference topological relation based on a deep feature extraction mechanism of a convolutional neural network to obtain a subdivided topological structure; Deep extraction and quantitative analysis of wound deformation characteristics are carried out on the subdivided topological structure, characteristic information of small deformation and small topological structure of the wound is accurately captured, and the proportional calibration parameters are obtained through calculation by combining the characteristic information obtained through extraction and analysis; Performing geometric pose normalization and scale standardization treatment on the wound three-dimensional model according to the proportion calibration parameters to obtain a treated wound three-dimensional model; According to the quantized and marked three-dimensional model, performing multi-view standardized shooting under a world coordinate system through a virtual camera to obtain a two-dimensional image containing wound morphology, a physical reference point and quantized and marked information; Inputting the two-dimensional image and the associated quantized labeling data into a visual language large model subjected to legal medical expert injury data fine adjustment, extracting visual features and quantized features of wounds, and completing judgment of a wound formation time interval and identification of organic injury based on the extracted features. Further, an original three-dimensional model of the wound site is acquired using a structured light scanner and is preprocessed to eliminate scanning noise and repair data defects, thereby obtaining a standardized three-dimensional model of the wound, comprising: Receiving an original three-dimensional model uploaded by a structured light scanner, and separating and extracting point cloud data and an initial grid model in the original three-dimensional model; Calculating the distance mean value and standard deviation of each cloud point in the neighborhood range by performing statistical filtering processing on the point cloud data, and removing outliers with deviation mean values