Bilkent University
Department of Computer Engineering


Real-Time Gesture Recognition for Cerebral Palsy Occupational Therapy Exercises


Mehmet Faruk Ongun
MSc Student
Computer Engineering Department
Bilkent University

Gesture Recognition is one of the widely researched subjects as of late. Besides the necessity and benefits of this area, the emergence of depth cameras/sensors has heavily contributed to this popularity because of the skeleton data provided by these cameras.

Using gesture recognition together with depth cameras can have many application areas. Fitness exercises is one of these areas and is being utilized concurrently. In our research, we are focusing on recognizing and determining the correctness of certain exercises and movements (specifically, occupational therapy exercises) for children with cerebral palsy using Hidden Markov models.(HMMs) The reason we chose HMM algorithm is the wide usage of this algorithm in gesture recognition applications and also the need of being able to learn new movements that can be needed by the occupational therapists.

The difference of our objective from standard gesture recognition is that in the treatment process of these children, very slight differences in movements may need to be detected in some cases and this cannot be achieved by using simple gesture recognition methods in real time. Another difficulty comes from the nature of HMM algorithm; HMM choses the closest gesture model to the test data even when the test data is a completely irrelevant one because a threshold cannot be applied. As a solution to these problems, we propose a system that uses HMM algorithm with garbage models which continuously modifies the models while application is being used by the patients(after the training process) and incorporates Dynamic Time Warping to improve performance.


DATE: 06 April, 2015, Monday @ 16:15