基于神经网络的三维人体建模方法

A 3D Human Body Modeling Method Based on Neural Network

利用扫描人体数据库,按特征点位置确定若干截面,对截面上的特征点重新识别和处理,获得胸部、脖子、腰部和臀部等特征部位截面环的训练样本,利用4层BP神经网络模型,通过编程分别训练得到人体胸部、颈部、腰部和臀部的神经网络权值参数,从而得到适用于一般人体的截面环自动生成模型,自动生成用户人体各个特征部位的截面环,最终生成三维人体网格模型.结果表明:该方法能很好地逼真实际人体特征曲线,是在只有用户人体参数信息,无扫描人体情况下的一种较好的人体建模方法.

A method of 3D human body modeling was presented. The model of four-layer feed forward BP （back-propagation） neural network was given. By making use of scanned body data, cutting planes were allocated to get the feature points of section contours of scanned body, and the training samples of the neck, bust, waist and hip were obtained. The weight and the feature curve modeling of the neck, bust, waist and hip can be got respectively after training. Based on section loops which generate from neural network, a 3D manikin was regenerated. Finally, the error analysis was done and the results show that the method can approach the real human body feature curve. It is an effective modeling way for the human body.