Multi-View Action Recognition Method Based on Regularized Extreme Learning Machine

In order to overcome the limitations of a single view of action recognition by using multi-view information, this paper proposes a novel method of multi-view human action recognition. The main steps involved in the process of recognition of action features are represented by fuzzy vector quantization; the membership vector is obtained after the action feature fuzzy. A fast and efficient Extreme Learning Machine (ELM) training algorithm based on single hidden layer feed-forward neural networks is proposed. Through the appropriate optimization of extreme learning machine, we reduce the dimension of the new feature space, and make full use of labeled and unlabeled examples to improve the action recognition accuracy. We evaluate our approach on both KTH and UCF50 action recognition databases, the results show that the recognition effect is better.