Abstract: In constructing 3D models of female body based on 2D photos, the complexity and diversity of the waist-abdomen-hip region should be taken into account, which is important for the accuracy of a reconstructed 3D model. Therefore, a piecewise fitting mapping model construction method is proposed in this paper to improve the accuracy of the mapping model between 2D contour features and 3D feature indexes.
In this paper, 3D point cloud data of 120 young women aged 18‒25 years were collected, and 15-layer feature sections and radial distance were extracted as 3D feature parameters. At the same time, six contour feature parameters were extracted from the front and side photos of human body. On the basis of correlation analysis, a piecewise mapping model was established to realize the mapping from 2D contour feature to 3D feature index. Then, trigonometric function was used to convert radial distance into discrete point cloud data. Finally, the idea of point-line-plane modeling was introduced to build the 3D model of waist, abdomen and hip.
In this paper, the model reconstruction error was used as the evaluation criterion to determine the optimal configuration of the layer number of feature section. When the number of feature layers was 15, the reconstruction error was not only reduced to 6.45 mm, but also the balance between the model complexity and the error value was achieved. In view of the diversity and complexity of female waist, abdomen and hip, this paper analyzed the correlation between the radial distance and the corresponding width and thickness, aiming to further improve the accuracy of the prediction model. In order to enhance the correlation of some feature points, the method of piecewise mapping was adopted in this paper, and the correlation coefficient of more than half of the samples was raised to the level of more than 0.85. On this basis, 100 samples were randomly selected as training sets, and a regression mapping model with width and thickness as independent variables and radial distance as dependent variables was constructed. To verify the prediction accuracy of the model, the remaining 20 samples were used as test sets for verification. The results show that the error of the mapping model is controlled within -1.5‒2.5 mm. Further, this paper used trigonometric function to carry out point cloud processing of radial distance set, and followed the idea of point-line-plane modeling to realize the reconstruction of female waist-abdomen-hip 3D model. By comparing the reconstructed model with the actual model, it is found that the maximum error percentage of the proposed method is limited to less than 3%.
The 3D human body modeling method based on 2D photo feature point mapping is particularly effective for complex female body types, and its core lies in the accurate positioning and mapping of feature points to build high-precision 3D models. The piecewise fitting technique further optimizes this process and significantly improves the correlation between 2D profile and 3D section parameters. Segmented mapping can refine the modeling for different parts of the body and improve the reconstruction accuracy. 

Key words: waist-abdomen-hip, 2D photo contour features, 3D section feature, mapping model, point-line-plane modeling

摘要： 为实现二维照片特征点重构腰腹臀三维模型，提出分段拟合的映射模型构建方法。首先获取三维点云及正侧面照片信息，探究特征截面层数并提取射线距离表征三维截面特征，通过图像处理方法获取正侧面照片腰腹臀的宽度及厚度6个尺寸表征二维照片轮廓特征。采用皮尔逊相关系数，结合分段策略确定宽度、厚度与射线距离的相关性，从而建立二者之间的映射模型。最后通过点坐标求解将射线距离集坐标化,采用点—线—面的建模方法，实现二维照片特征点映射的腰腹臀部三维模型构建。结果表明：15层特征截面能有效表征腰腹臀部形态，重构误差达6.45 mm。相较于整体拟合，分段策略能显著提升相关性，75.9%样本相关性达到0.85以上，映射模型误差在-1.5~2.5 mm之间。通过重构与实际模型的尺寸对比，发现模型重构误差在3%以内，证明该构建方法的可行性，这可为二维照片轮廓重构三维人体模型提供参考。

关键词: 腰腹臀部, 二维照片轮廓特征, 三维截面特征, 映射模型, 点—线—面建模