Human Action Recognition using Spatial-Temporal Analysis and Bag of Visual Words

Human Action Recognition (or HAR) is gaining interest in a wide range of domains from domestic activities to industrial environments. HAR refers to the ability of a computerized system to correctly identify and analyze human activities and behaviors, equivalent to what is known as perception to humans. The concept lends itself to a variety of applications which include surveillance, entertainment and the monitoring of the elderly. Various techniques and approaches exist in the literature through which HAR has been achieved, this paper focuses on Motion History Images and its variants. The proposed method generates Motion History Images (MHI) and extracts features using the Bag of features approach for training. The bag of features approach extracts Speed up Robust Features (SURF) and then clusters them using k-means clustering to form a training vector. The training vectors obtained are then trained using support vector machines (SVM). The performance of the proposed method is evaluated using the Weizmann and the KTH datasets.