In human-robot interaction, gestures play an important role in transferring messages from the human to the robot and vice versa. The robot has to be able to properly understand the gestures in order to respond in an appropriate manner. This research work focuses on dynamic gestures' recognition using robot's cameras, which capture the performed gestures in a certain interactional scenario. The gestures are processed and classified to a pre-defined category of gestures. Our gesture recognition system is currently part of a robot behavior control architecture developed for children with autism.