Human Gait Behavior Classification using HMM based on Lower Body Triangular Joint Features

Myagmarbayar Nergui, Nevrez Imamoglu, Yuki Yoshida, and Wenwei Yu


Human behaviour classification, HMM, NN, ANN, mobile robot, home healthcare


The goal of our research is to develop an autonomous at home health care mobile robot for motor-function impaired patients (MIPs). This paper focuses on one of the key component, human gait behaviour classification, based on the side view observation. The purpose is to identify gestures such as walking, impaired walking, standing, and sitting where the impaired walking is the key gesture to be classified for future analysis of the patient’s walking condition during their daily walking practices for the rehabilitation process. For this purpose, we have proposed new method of using triangular representation with three colour markers attached to ankles and hip joints. Internal angles of triangle can be calculated from side viewpoint tracking to create the feature space for classification based on the Hidden Markov Model (HMM). Then, the proposed HMM based classification is compared to the Nearest Neighbour (NN) and Artificial Neural Networks (ANN) classification methods. Experimental results showed that human gait behaviour classification employing HMM can be achieved from the joint trajectory of triangle with high accuracy by outperforming the compared classification methods.

Important Links:

Go Back