摘要
针对单幅图像资料下的三维人物模型网格模型动态拟合任务,使用阵列归一化算法对位图图像进行预处理,使用循环多列神经网络对拟合网格进行深度赋值,最终通过IoU评价体系和CD评价体系对训练过程进行辅助评价,使用标准偏差法对实际运行阶段图像进行辅助评价。实验结果一是革新算法收敛速度弱于传统算法;二是革新算法在IoU指标和CD指标上的提升比例超过上述4个传统算法。革新算法调用的神经网络节点量较高。结果表明,基于单幅图像获取三维动画模型的算法模型具有较强可用性。
Aiming at the dynamic fitting task of 3 D human model mesh model under single image data, the array normalization algorithm is used to preprocess the bitmap image, and the cyclic multi column neural network is used to assign the depth of the fitting grid. Finally, the training process is evaluated by the IoU evaluation system and the CD evaluation system, and the standard deviation method is used to assist the image in the actual operation stage Evaluation. The results show that the convergence speed of the improved algorithm is weaker than that of the traditional algorithm, and the promotion ratio of the improved algorithm in IoU index and CD index is higher than the above four traditional algorithms. The number of neural network nodes called by the innovation algorithm is high. The results show that the algorithm model of 3 D animation model based on single image has strong usability.
作者
樊鸿涛
Fan Hongtao(Shangluo University,Shangluo 726000,China)
出处
《电子测量技术》
2020年第24期99-102,共4页
Electronic Measurement Technology
关键词
单幅图像
三维拟合
人物动画
三维重建
仿真验证
single image
3D fitting
character animation
3D reconstruction
simulation verification
