Action Recognition Using Local Joints Structure and Histograms of 3D Joints

In this paper, we present a method for human action recognition using local joints structure and histograms of 3D joints. Global features like histograms of 3D joints [12] ignore the local structure information of the human body joints, which is also essential for accurate action recognition. To address this problem, we propose a local joints structure feature as a complement, and combine both global and local features for posture description in our method. Then, linear discriminant analysis is used to reduce the feature dimension, and k-means clustering is utilized to generate codewords. Finally, these codewords are treated as discrete symbols for training hidden Markov models (HMMs) which are used for action recognition. Experimental results demonstrate that our method has better performance than other methods when testing on UTKinect-Action Dataset and MSR Action3D dataset.