基于特征层融合和随机投影的行为识别算法

Human Action Recognition Based on Feature Level Fusion and Random Projection

提出一种基于特征层融合和随机投影的行为识别算法;该方法提取视频序列的时空梯度特征和Gabor特征;然后进行特征层融合,得到分类能力更强的特征,有效地表征人体行为;同时,使用随机投影对融合后的特征进行降维;最后,为了解决主题模型参数估计迭代复杂的问题,将贝叶斯参数估计法应用于LDA(latent dirichlet allocation)主题模型中,对视频中的行为进行分类。在公开的KTH和Weizmann数据集上进行了实验,结果表明方法不仅比单一局部时空特征描述符识别性能好,而且在相同实验设置下,也优于其他基本分类器。

A novel method for human action recognition based on feature level fusion and random projection is proposed. The proposed method exploits both spatial-temporal gradient features and Gabor features of the action in video. This helps representing the action more accurately after feature level fusion. Meanwhile,the random projection is employed to reduce the dimensionality of features effectively. Finally,in order to deal with the complex iteration issue in the process of parameter estimation,the Bayesian parameter estimation is applied to the Latent Dirichlet Allocation( LDA) topic model. Experimental results on publically available datasets KTH and Weizmann dataset demonstrate that the proposed method not only achieves the better performance than the single local descriptor approach but also improves the recognition performance compared with the baseline classifier in the same experimental settings.