Trajectory classification using HMMs

Abstract

This paper focuses on evaluation of motion of objects through classification of their trajectories. The objects used for evaluation of the presented method are objects tracked in video sequence but the method is quite general and can be used for more general trajectories. The main potential application of the presented method is detection of abnormal behavior of humans, hand gesture detection and recognition, etc. Hidden Markov Models (HMMs) and Gaussian Mixture Models are used as the base of the the classification mechanism. The paper contains description of the method, description of experiments and their results, and conclusions.