Abstract
This paper presents a novel method to the analysis of human-arm motion, in particular improving the efficiency of conventional motion recognition algorithms. Contrary to the prior art methods, this research develops a framework for human-arm motion recognition where qualitative normalised templates (QNTs) is proposed to replace the conventional approaches. First of all, the conventional robotic model has been employed to build a generic vision model for a human-arm, that is we utilise the robot kinematics to construct a stick model. Secondly, the qualitative robotic model is adopted to learn and construct the QNTs where human-arm motion is termed as, whose execution is consistent and could be easily characterised by a definite space-time trajectory in configuration space. Finally, classification of the human-arm motion is achieved by comparing the QNTs to the parameters learnt with particle filter based motion tracking algorithm. Experimental evaluation has demonstrated the effectiveness of the proposed method in human-arm motion classification, and our future work is focused on extending the proposed method to recognise complex human motion, e.g. walking and running.