No abstract

Nowadays, human activity recognition(HAR) becomes a hot topic with broad applications. Some researches have conducted HAR from microscopic perspective and achieved good results. In this article, two methods are proposed for further improvement. Firstly, an improved symbolization method with stacked sparse autoencoder is proposed for better data symbolization. Secondly, an improved multi-classification Adaboost is proposed to further optimize the recognition effect, and it is more suitable for the application scenario of this article. In the experiments section, firstly, e xperiments a nd a nalysis about various influencing p arameters a re c onducted, t hen comparison experiments with several new or representative methods are carried out, and finally five representative sensor activity datasets(UCI Sports and Daily dataset, Wisdm Phoneacc&Watchacc dataset, Skoda dataset, HAPT dataset) are used to prove the universal applicability and achieve satisfactory effect.

No abstract