An improved geometric descriptor associated with wavelet transform for aggressive human behaviour recognition

摘要

Actually, the automatic recognition of aggressive human behaviours becomes an important issue to improve the intelligent security systems and enhance the public safety. In this paper, we propose a robust algorithm which aims to detect an aggressive human behaviour from a monocular vision. The proposed algorithm is based on a spatio-temporal descriptor by using a geometric approach. The latter describes the spatial information of the specific limbs of the body such as arm limb in the form of signatures. To extract the spatio-temporal features, the signatures are normalised and then analysed by tbe one-dimensional discrete wavelet transform. Thus, a binary support vector machine classifier is used in order to separate either aggressive or non-aggressive behaviours. Several tests have been conducted on the KTH dataset. The obtained results show that the proposed method enables robust recognition in challenging situations, and it is suitable in online action recognition with significant accuracy rate.