Learning Principal Orientations Descriptor for Action Recognition

In this paper, we propose an unsupervised learning based representation method, named principal orientations descriptor (POD), to describe the local and statistic characteristics for action recognition. Unlike hand-crafted features which require high prior knowledge, our POD is learned from raw pixels and reflects the distribution of principal orientations around the motion trajectories. Different from deep learning based features which are based on a large number of labeled data, our POD is learned in an unsupervised learning manner. We learn POD in the spatial domain and the temporal domain based on the same motion trajectories individually, which makes POD have the ability to describe both the spatial and the temporal information along the same trajectories. To evaluate the performance of POD, we conduct experiments on two challenging action datasets: Hollywood2 and HMDB51. The results show that our method is competitive to the existing methods.