Research On Self-paced Learning Algorithm For Human Action Recognition

Nowadays,the application of human action recognition in video can be seen everywhere in people's daily life,such as intelligent monitoring,video retrieval,virtual reality and so on.Therefore,video-based human action recognition has very broad development prospect and high research value.In this paper,three kinds of recognition algorithms are proposed:human action recognition based on multi-view self-paced learning,human action recognition based on multi-view curriculum sequence of multi-task learning,human action recognition based on multi-view self-paced learning sequence of multi-task learning.First,this paper has carried on the thorough research to the novel and unique self-paced learning algorithm,and proposed a algorithm of human action recognition based on multi-view self-paced learning.In addition to preserving the excellent characteristics of self-paced learning itself,the algorithm also considers the impact of different views on the courses sequence of self-paced learning.This algorithm integrates self-paced learning classifiers from multiple views by using linear programming boosting algorithm,and further improves the recognition effect by learning a more comprehensive course which can solve the problems of complex human action recognition.Then,this paper applies curriculum learning of multiple tasks to human action recognition for the first time.This method integrates the thought of curriculum learning into multi-task learning,and designs the course sequence according to the degree of difficulty and relevance of the tasks.On this basis,this paper proposes a multi-task learning algorithm based on multi-view course sequence.This algorithm learns multi-task courses from multiple views,and integrates curriculum learning of multi-task models from multiple views through linear programming boosting algorithm to obtain a more complete course sequence.In the end,this paper proposes a multi-task learning algorithm based on multi-view and self-paced learning sequence.This algorithm not only takes into account the curriculum design of multiple tasks,but also embeds the curriculum of each task into the model by self-paced learning algorithm.By designing and fusing the courses under different views,this method can get a more comprehensive curriculum model.The experimental results on the open data set show that the algorithms proposed in this paper are more robust than the original algorithms,and the recognition effect is better.