Publication

Abstract : Human Action Recognition (HAR) has proven to be an important field of research in data vision, which has applications scattered in monitoring, health care, sports analysis and human computer contact. Despite significant progress, there are challenges in identifying tasks in different environments, attitudes and styles. This article introduces a new framework for human action recognition improved by the paper transmission, which benefits from the strength of the action recognition model and the transfer of nerve style to improve generality. By moving stylistic functions from one domain to another, our approach makes the model learning the irreversible representation of tasks, reduces the effect of domain changes and improves performance in unavailed scenarios. We evaluate our method on the reference date set, and demonstrate its effectiveness in identifying tasks under different circumstances, such as changes in light, background and movement styles. The results suggest that each framework our style transmission-aggression does better by traditional methods, and provides a promising direction for future research in recognition across domains. This work not only moves as a condition -by -species in has, but also opens new ways to integrate artistic and stylistic elements into the data viewing system.