Behavior Recognition based on Signal Processing Technology

DOI:10.21742/ijhit.2653-309X.2021.1.2.05

AUTHORS

Mohamed Osama,Southern Cross University, Bilinga, Australia

ABSTRACT

With the continuous development of sensor technology, more and more types of data are applied to the research of behavior recognition. A single type of data can only describe the current behavior of the target one-sidedly while adding different types of data can add multi-dimensional description supplement to the target behavior. To get higher recognition accuracy, multimodal behavior recognition technology using or integrating multi-source data features has gradually developed into a new research topic. In this paper, a multimodal system composed of three data sources: brainwave signal, image sequence, and motion signal is used to study behavior recognition. Signal acquisition and preprocessing feature extraction, and modeling recognition are the three essential processes of classic behavior recognition research. In modeling recognition, this paper combines SVM, random forest, XGBoost, and Light GBM with different feature combinations, and models individual data sets with a single algorithm, multi-person data sets with a single algorithm, and integrated algorithm respectively. The final experimental results show that multi-source signals can achieve a better recognition effect than a single source. For the feature extraction method of EEG signal, the new method proposed is superior to the method based on traditional filtering. The indicators based on the ensemble model are better than those based on the single algorithm model, and the optimal classification accuracy is over 90% for both the single algorithm model and the ensemble model.