A comparison of SVM and RVM for human action recognition

Human action recognition is a task of analyzing human action that occurs in a video. This paper investigates action recognition by using two classification techniques, namely Relevance Vector Machine (RVM) and Support Vector Machine (SVM). SVM is a technique for supervised classification that used in statistics and machine learning. By separating the distinct class with a maximum possible wide gap, SVM tries to predict the respective class given a set of input data. On the other hand, RVM is a Bayesian model of Generalized Linear Model (GLM) that has an identical function with SVM. RVM uses significantly fewer basis functions as it uses Bayesian inference with a prior distribution on weight thus makes solution sparse. Experimental studies on a human action dataset show that RVM is better as compared to SVM on action recognition. Although RVM takes more training time, however, it requires fewer testing time than SVM. RVM model is more general because it contains minimum basis function. Therefore, it is more robust compared to SVM. RVM performs good classification on action recognition that contains large dataset.