Human activity recognition via temporal moment invariants

Temporal invariant shape moments intuitively seem to provide an important visual cue to human activity recognition in video sequences. In this paper, an SVM based method for human activity recognition is introduced. With this method, the feature extraction is carried out based on a small number of computationally-cheap invariant shape moments. When tested on the popular KTH action dataset, the obtained results are promising and compare favorably with that reported in the literature. Furthermore our proposed method achieves real-time performance, and thus can provide latency guarantees to real-time applications and embedded systems.