engleski

Application of Hidden Markov Models for Gesture Recognition in Functional and Symbolic Imitation

This work presents implementation of evaluation software for interaction between a NAO robot and a child in robot assisted autism diagnostic. Evaluation of interaction between a robot and a child is done through the ADOS (Autism Diagnostic Observation Schedule) protocol. Interaction is based on a played scenario where the robot is performing a certain gesture and it observes the child’s response when the imitation of the same gesture is expected. Functional and symbolic imitation consists of several parts. This is why it is implemented as a finite state machine. The protocol starts by robot and program initialization which is followed by object detection, assessment of its shape and dimensions which is information crucial for object manipulation. The last part of the protocol is gesture recognition of the object trajectory using Hidden Markov Models(HMM). Kalman filter is implemented in order to achieve better gesture recognition. The result is an imitation protocol which is robust and very precise.Functionalities that enable evaluation of interaction are developed. They have provided good results in laboratory testing conditions which led to successful accomplishment of the first phase of robot and child interaction evaluation.

Autism; robot assisted diagnosis; Finite state machine; machine learning; Hidden Markov Models; Kalman filter; NAO robot

nije evidentirano