A semi-supervised learning system for service robots to recognise human actions

Human action recognition is an essential ability for service robots to interact with human users. This paper presents a novel semi-supervised learning system for a service robot to recognise human actions and provide services. The multi-class co-training algorithm is firstly adopted to leverage the information from unlabelled data using two different classifiers selected by means of the diversity measure of entropy. The confidence score is then derived from both the nearer labelled and the unlabelled neighbours. The unlabelled examples with both the higher and the lower confidence scores are added to the labelled training set. To evaluate the proposed algorithm, mixed descriptors are used to express actions so that the recognition algorithm can quickly complete the recognition process from a single frame of visual image. Experimental results are presented to show that the proposed 3D vision-based semi-supervised learning system can recognise simple human actions effectively. © 2014 Taylor & Francis and The Robotics Society of Japan.