基于MIV特征筛选和BP神经网络的三维人体参数转换

Conversion of 3D Humanbody Parameters Based on MIV Algorithm Selection and BP Neural Network

在BP神经网络建模技术的基础上,提出引入平均影响值(MIV)算法筛选采集的变量参数,构建三层前馈神经网络求解三维人体建模参数的方法.结果表明,通过逐步删除后10%序位的参数变量,所建模型的最大误差为0.23,平均误差为0.058 667,训练精度为4E-3,预测精度为94.2%,可用于实际三维试衣时的参数计算.

Based on BP neural network modeling technology,a new approach of MIV algorithrm selection was applied to appraise 3D humanbody parameters and a three-layered feed forward neural network method was constucted for 3D humanbody parameters modeling. The results show that after removing 10% bit from the end of parameter variables sequence step by step,the models＇ errors are all lower than 0.23,the average error is about 0. 058 667, the training accuracy is 4E - 3, the prediction accuracy is 94.2% ,and which can be used in parameters calculation of 3D fitting in practice.