Abstract: The body's movement process is relatively complex,so there are many similar movements in the images,for-ming interferences on the characteristics of the traditional recognition,and causing low recognition accuracy．In order to improve its recognition accuracy,this paper put forward a kind of Hu moment invariants and artificial fish optimization support vector machine (SVM) model for human motion recognition (Hu-AFSA-SVM)．First of all,based on the two-dimensional continuous images,extracted the image 7Hu moment invariants of the human body movement gesture recognition,and then input into SVM to train,and picked a AFSA to SVM parameter optimization,to find an optimal hyperplane,in as much as possible meet the constraints of classification,all concentrated human motion data classification categories,complete recognition．Finally carried out the simulation experiment．Simulation results show that,compared with other identification model,Hu-AFSA-SVM improves the human motion recognition accuracy,speeds up the recognition, and is an effective method for human movement gesture recognition.

Key words: Human gesture,Support vector machine,Moment invariants,Artificial fish swarm algorithm,Recognition