CN-121999027-A - Human body size data determining method, device, medium and product based on 3D modeling

CN121999027ACN 121999027 ACN121999027 ACN 121999027ACN-121999027-A

Abstract

Human body size data determining method, device, medium and product based on 3D modeling. The method comprises the steps of collecting multi-view visible light images and multi-view depth maps of a target human body to generate multi-view point cloud data, carrying out alignment fusion on the multi-view point cloud data through iterative minimization of corresponding point distance deviation between the multi-view point cloud data to generate a global point cloud model, classifying the global point cloud model to obtain human body point clouds, carrying out grid reconstruction based on the human body point clouds to generate an initial three-dimensional grid model, carrying out smoothing treatment on the initial three-dimensional grid model to obtain a final three-dimensional grid model, positioning human skeleton datum points of the final three-dimensional grid model, and extracting human body size data. By implementing the technical scheme provided by the application, the technical problems that the accuracy of the extracted critical dimension data such as human body circumference, length and the like is insufficient and the requirement of a high-precision application scene is difficult to meet are effectively solved.

Inventors

WANG JINGWU
ZHUANG RUIQIANG

Assignees

山东圣梵尼服饰股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260120

Claims (10)

1. A human body size data determining method based on 3D modeling, characterized by being applied to a server, the method comprising: Synchronously acquiring a multi-view visible light image and a multi-view depth map of a target human body to generate multi-view point cloud data; Performing alignment fusion on the multi-view point cloud data by iteratively minimizing the corresponding point distance deviation between the multi-view point cloud data to generate a global point cloud model; Classifying the global point cloud model through a preset semantic segmentation network to distinguish human body parts from non-human body parts, and removing the non-human body parts to obtain human body point cloud; performing grid reconstruction based on the human body point cloud to generate an initial three-dimensional grid model containing preset geometrical characteristics of human body parts; Smoothing the initial three-dimensional grid model according to a preset topology protection mechanism to obtain a final three-dimensional grid model; And positioning human skeleton datum points of the final three-dimensional grid model by using a preset detection algorithm, and extracting human size data according to the human skeleton datum points.
2. The method of claim 1, wherein the synchronously acquiring the multi-view visible light image and the multi-view depth map of the target human body to generate multi-view point cloud data comprises: Disposing a plurality of acquisition units around a preset scanning area in a ring-shaped array mode, wherein each acquisition unit in the plurality of acquisition units comprises a visible light camera and a depth camera which are coaxially arranged; Simultaneously transmitting a hardware synchronous trigger signal to the plurality of acquisition units to acquire the multi-view visible light image and the multi-view depth map; and converting depth data in the multi-view depth map into three-dimensional coordinate points based on preset camera parameters, and adding color data in the multi-view visible light image as attributes to the corresponding three-dimensional coordinate points to generate multi-view point cloud data.
3. The method according to claim 2, wherein the converting depth data in the multi-view depth map into three-dimensional coordinate points based on preset camera parameters and adding color data in the multi-view visible light image as attributes to the corresponding three-dimensional coordinate points to generate the multi-view point cloud data includes: acquiring a depth value corresponding to each two-dimensional pixel coordinate in the multi-view depth map; converting the two-dimensional pixel coordinates and the depth values into three-dimensional coordinate points under a camera coordinate system based on an internal reference matrix in the camera parameters; And acquiring a color value corresponding to the two-dimensional pixel coordinate in the multi-view visible light image, and adding the color value to the three-dimensional coordinate point to generate multi-view point cloud data.
4. The method of claim 1, wherein classifying the global point cloud model through a preset semantic segmentation network to distinguish human body parts from non-human body parts and reject the non-human body parts to obtain human body point clouds comprises: determining a neighborhood point set for each point cloud coordinate in the global point cloud model; For each point cloud coordinate, aggregating, by a feature encoder in the semantic segmentation network, self features of the point cloud coordinate and geometric relationships of all points in the neighborhood point set corresponding to the point cloud coordinate to generate a context feature vector; According to each context feature vector, calculating to obtain a human semantic category score and a non-human semantic category score for each point cloud coordinate through a preset classification algorithm; And judging the semantic category of each point cloud coordinate according to the human semantic category score and the non-human semantic category score, and constructing the point cloud coordinate with all the semantic categories being human parts as the human point cloud.
5. The method according to claim 1, wherein the smoothing the initial three-dimensional mesh model according to a preset topology protection mechanism to obtain a final three-dimensional mesh model includes: In a preset iterative smoothing period, performing smoothing iteration on the initial three-dimensional grid model; In each smoothing iteration, calculating the coordinate to be updated of each vertex according to the adjacent vertex coordinates of each vertex in the initial three-dimensional grid model; Updating the coordinates of all the vertexes of the initial three-dimensional grid model to the corresponding coordinates to be updated to obtain a temporary updated grid; Calculating a topology difference value between the temporary updated grid and the initial three-dimensional grid model, and comparing the topology difference value with a preset topology zero threshold; If the topology difference value is equal to the topology zero threshold, the temporary updating grid is used as the input of the next smoothing iteration, and if the topology difference value is not equal to the topology zero threshold, the initial three-dimensional grid model is used as the input of the next smoothing iteration; and after the iterative smoothing period is finished, obtaining the final three-dimensional grid model.
6. The method of claim 1, wherein said locating human skeletal fiducials of the final three-dimensional mesh model using a predetermined detection algorithm comprises: calculating a geometric feature vector based on the three-dimensional coordinates of each vertex of the final three-dimensional mesh model and the three-dimensional coordinates of the adjacent vertex of each vertex; Inputting all the geometric feature vectors into a preset geometric depth learning network, and generating a corresponding probability heat map for each preset human skeleton datum point type; in the probability heat map corresponding to each of the human skeleton reference point types, a target vertex having a maximum probability value is positioned, and three-dimensional coordinates of the target vertex are determined as the human skeleton reference point.
7. The method of claim 1, wherein the extracting human dimension data from the human skeletal reference points comprises: Combining the positioned human skeleton datum points into a plurality of measuring point pairs according to a preset human body size definition library, wherein each measuring point pair corresponds to a human body part to be measured; For each of the measurement point pairs, calculating a shortest geodesic distance between two of the human skeletal fiducials along the surface of the final three-dimensional grid model; and taking the calculated shortest ground wire distance as a size value corresponding to the human body part to be measured, and summarizing all the size values to generate the human body size data.
8. An electronic device comprising a processor, a memory, a user interface, and a network interface, the memory for storing instructions, the user interface and the network interface each for communicating with other devices, the processor for executing instructions stored in the memory to cause the electronic device to perform the method of any of claims 1-7.
9. A computer readable storage medium storing instructions which, when executed, perform the method of any one of claims 1-7.
10. A computer program product, characterized in that the computer program product, when run on an electronic device, causes the electronic device to perform the method of any of claims 1-7.

Description

Human body size data determining method, device, medium and product based on 3D modeling Technical Field The application relates to the technical field of computer vision and three-dimensional modeling, in particular to a human body size data determining method, device, medium and product based on 3D modeling. Background Along with the development of fields such as virtual fitting and personalized clothing customization, the acquisition of an accurate three-dimensional model of a human body becomes a key technology. The current mainstream human body three-dimensional modeling technology generally utilizes sensor arrays deployed in a plurality of directions to synchronously capture the space information of the human body surface, so that the information from different view angles is registered and combined under a unified coordinate system to construct a complete human body point cloud, and finally, the point cloud data is converted into a three-dimensional digital model with a continuous surface through gridding processing. However, the three-dimensional grid model surface directly reconstructed from the original scan data often contains noise and irregular fluctuations, and these flaws can interfere with the subsequent human body dimension measurement process, so that the accuracy of the finally extracted critical dimension data such as human body circumference, length and the like is insufficient, and the requirements of high-precision application scenes are difficult to meet. Disclosure of Invention In order to solve the technical problems, the application provides a human body size data determining method, device, medium and product based on 3D modeling. In a first aspect of the present application, a method for determining human body size data based on 3D modeling is provided, and the following technical scheme is adopted: Synchronously acquiring a multi-view visible light image and a multi-view depth map of a target human body to generate multi-view point cloud data; Performing alignment fusion on the multi-view point cloud data by iteratively minimizing the corresponding point distance deviation between the multi-view point cloud data to generate a global point cloud model; Classifying the global point cloud model through a preset semantic segmentation network to distinguish human body parts from non-human body parts, and removing the non-human body parts to obtain human body point cloud; performing grid reconstruction based on the human body point cloud to generate an initial three-dimensional grid model containing preset geometrical characteristics of human body parts; Smoothing the initial three-dimensional grid model according to a preset topology protection mechanism to obtain a final three-dimensional grid model; And positioning human skeleton datum points of the final three-dimensional grid model by using a preset detection algorithm, and extracting human size data according to the human skeleton datum points. By adopting the technical scheme, the multi-view visible light image and the depth map are synchronously acquired, point cloud data are iteratively aligned and fused, a high-precision global point cloud model is constructed, human point cloud purity is ensured by intelligently rejecting non-human interference through a semantic segmentation network, a three-dimensional grid model retaining key geometric features is generated through grid reconstruction and topology protection smoothing processing, and finally human size data are automatically extracted based on skeleton datum points. The method effectively solves the technical problems of difficult multi-view data fusion, large interference of non-human body parts, insufficient smoothness of the model, inaccurate size extraction and the like in the traditional human body size measurement, and realizes non-contact, automatic and high-precision human body size determination. Optionally, the synchronously collecting the multi-view visible light image and the multi-view depth map of the target human body to generate multi-view point cloud data includes: Disposing a plurality of acquisition units around a preset scanning area in a ring-shaped array mode, wherein each acquisition unit in the plurality of acquisition units comprises a visible light camera and a depth camera which are coaxially arranged; Simultaneously transmitting a hardware synchronous trigger signal to the plurality of acquisition units to acquire the multi-view visible light image and the multi-view depth map; and converting depth data in the multi-view depth map into three-dimensional coordinate points based on preset camera parameters, and adding color data in the multi-view visible light image as attributes to the corresponding three-dimensional coordinate points to generate multi-view point cloud data. By adopting the technical scheme, the strict space-time alignment of the multi-view images and the depth data is realized, and the problem of data dislocation caused by vi