A Novel Action Recognition Method Based on Improved Spatio-Temporal Features and AdaBoost-SVM Classifiers

Abstract

Most of existed action recognition methods based on spatio-temporal descriptors have ignored their spatial distribution information. However the spatial distribution information usually is very useful to improve the discriminative ability of the motion representation. An improved spatio-temporal is proposed in this paper by combining local spatio-temporal feature and global positional distribution information (FEA) of interest points. Furthermore, in order to improve the classifier’s performance, an Adaboost-SVM method is utilized to recognize the human actions by using the proposed motion descriptor. The proposed recognition method is tested on the public dataset of KTH. The test results verified the proposed representation and recognition method can more accurately describe and recognize the human motion.