An Approach to Detecting Person's Unwonted Behavior for Service Robot

Detecting the person's unwonted behavior state timely and correctly in home environment for the service robot is a fundamental problem. This paper proposes an approach to this problem using the position and pose information as the features and the computing result of the CRF model as the judging gist. Firstly, a pose template based on position label is defined to reduce the dimension of the problem. Then, a space-time constrained CRF model is proposed by integrating the position information and pose features. Experiment results demonstrate that the model can run on a PC with high accuracy and is feasible to afford a judging gist to detect persons' unwonted behavior for service robot.