Class-Constrained Transfer LDA for Cross-View Action Recognition in Internet of Things
Shuang Liu · IEEE Internet of Things Journal · 2017
Internet of Things (IoT) is a fast-growing field based on different techniques and applications. In this paper, we focus on IoT in monitoring mission which is widely used in everyday life, such as security surveillance, health-care, independent living, etc. In order to overcome the challenges of various viewpoints and heterogenous sensors in IoT, we propose a novel method named class-constrained transfer linear discriminant analysis (CTLDA), which learns two projection matrices with different dimensionality for mapping the original features into a common subspace. The target of learning two projection matrices is to maximize the interclass data and minimize the intraclass data. Meanwhile, we propose the class-constrained regularization which enforces the neighboring samples with the same class to be still close to each other so as to further improve the discrimination of CTLDA. The class-constrained regularization possesses two strategies, i.e., hard constraint and soft constraint. The experimental results demonstrate that our method achieves better performance than the state-of-the-art methods.