Human Activity Recognition using Motion Feature and Two Stage Classification

Automatic analysis of the ongoing video to understand what is happening in the monitored area is of great practical use in many applications. Understanding the human activities from the videos is useful in applications like video surveillance, patent monitoring, content based retrieval etc. Proper selection of features plays important role in the performance of activity recognition system. The goal of this paper is to investigate the features to describe human activities and use of this to improve the recognition rate.  A novel motion feature and two stage classification is suggested in this paper. Four classifiers namely KNN, SVM, NN and NB are used for classification. The extracted features are represented using two subspace PCA and LDA. Observations show that use of motion feature improves the recognition rate for all the classifier.