基于高斯增量降维与流形Boltzmann优化的人体运动形态估计

发布时间：2018-02-28 09:22

本文关键词： 高斯增量降维模型流形Boltzmann优化人体运动形态轮廓图像子向量　出处：《电子学报》2017年12期 　论文类型：期刊论文

【摘要】：为了从多视角轮廓图像估计出含空间位置信息的三维人体运动形态,该文提出高斯增量降维与流形Boltzmann优化(GIDRMBO)算法.该算法把表示三维人体运动形态的高维数据分成表示空间位置信息和姿态信息两段子向量后,用高斯增量降维模型(GIDRM)分别对其样本进行降维,建立相应的低维空间及映射关系,然后在相应的低维空间使用流形Boltzmann优化算法来对轮廓匹配目标函数进行优化,从而实现估计.其中,所提算法分别利用了两段子向量样本的低维数据作为先验信息,可较好的避免陷入局部最优区域进行搜索,最终生成与各视角原始运动图像匹配且含空间位置信息的三维人体运动形态.经仿真实验验证,所提算法与常用粒子滤波算法相比,其估计误差小,并且还能起到消除轮廓数据歧义和克服短时遮挡的作用.
[Abstract]:In order to estimate three-dimensional human motion form with spatial position information from multi-view contour image, In this paper, Gao Si incremental dimensionality reduction and manifold Boltzmann optimization algorithm are proposed, in which the high dimensional data representing three-dimensional human motion is divided into two subvectors: spatial position information and attitude information. Using Gao Si's incremental dimensionality reduction model (GIDRM) to reduce the dimension of the sample, the corresponding low-dimensional space and mapping relationship are established, and then the contour matching objective function is optimized by using the manifold Boltzmann optimization algorithm in the corresponding low-dimensional space. In order to realize the estimation, the proposed algorithm uses the low-dimensional data of two subvector samples as prior information, which can avoid falling into the local optimal region to search. Finally, the 3D human body motion shape, which is matched with the original motion image of each angle of view and containing the spatial position information, is generated. The simulation results show that the proposed algorithm has less estimation error than the usual particle filter algorithm. It can also play the role of eliminating the ambiguity of contour data and overcoming the short-term occlusion.
【作者单位】：华南理工大学电子与信息学院;广东第二师范学院计算机科学系;
【基金】：国家自然科学基金(No.61202292) 广东省自然科学基金(No.9151064101000037) 广东省普通高校青年创新人才项目(No.2016KQNCX111)
【分类号】：TP391.41

【相似文献】