In this paper we propose a reliable gesture recognition system that could be run on low-level machines in real-time, which is practical in human-robot interaction scenarios. The system is based on a Random Forest classifier fed with Motion History Images(MHI) as classi?cation features. To detect fast continuous gestures as well as to improve the robustness, we introduce a feedback mechanism for parameter tuning. We applied the system as a component in the child-robot imitation game of ALIZ-E project.