融合生成模型和判别模型的双层RBM运动捕获数据语义识别算法

发布时间：2019-04-24 13:43

【摘要】：对人体运动捕获数据底层特征和高层语义之间常常存在语义鸿沟的问题,结合深度学习思想,提出一种融合受限玻尔兹曼机生成模型和判别模型的运动捕获数据语义识别算法.该算法采用双层受限玻尔兹曼机,分别对运动捕获数据进行判别性特征提取(特征提取层)和风格识别(语义判别层),首先考虑到自回归模型对时序信息具有出色的表达能力,构建一种基于单通道三元因子交互的条件限制玻尔兹曼机生成模型,用于提取运动捕捉数据的时空特征信息;然后将提取出的特征与对应的风格标签相耦合,作为语义判别层中受限玻尔兹曼机判别模型的当前帧数据层输入,进行单帧风格识别的训练;最后在获得各帧参数的基础上,在模型顶部加入投票空间实现对运动捕捉序列的风格语义的有效识别.实验结果表明,文中算法具有良好的鲁棒性和可扩展性,能够满足多样化运动序列识别的需求,便于数据的有效重用.
[Abstract]:In view of the problem that there is often semantic gap between the bottom features and the high-level semantics of human motion capture data, a semantic recognition algorithm for motion capture data is proposed, which integrates the constrained Boltzmann machine generation model and the discriminant model, combined with the idea of in-depth learning. The two-layer constrained Boltzmann machine is used to extract discriminative features (feature extraction layer) and style recognition (semantic discrimination layer) of motion capture data respectively. Firstly, considering that autoregressive model has excellent ability to express time series information, a conditional constrained Boltzmann machine generation model based on the interaction of single channel and three factors is constructed to extract spatio-temporal feature information of motion capture data. Then, the extracted feature is coupled with the corresponding style tag, which is used as the input of the current frame data layer of the restricted Boltzmann machine discriminant model in the semantic discrimination layer, and the training of the single frame style recognition is carried out. Finally, on the basis of obtaining the parameters of each frame, the voting space is added to the top of the model to realize the effective recognition of the style semantics of motion capture sequences. The experimental results show that the proposed algorithm has good robustness and extensibility and can meet the requirements of diverse motion sequence recognition and facilitate the effective reuse of data.
【作者单位】：华侨大学计算机科学与技术学院;厦门市模式识别与计算机视觉重点实验室;
【基金】：国家自然科学基金(61673185,61673186) 福建省自然科学基金(2015J01656) 华侨大学科研创新能力培养资助项目(1511414012)
【分类号】：TP391.41;TP391.9

【相似文献】