In imitation learning processes, the ”student” robot must be able to perceive the environment and to detect one ”teacher”. In our approach of learning by imitation, we consider that the student tries to learn the teacher trajectory (temporal pattern). In this context, we propose a neural architecture for a mobile robot which detects its teacher using the optical flow information. The detected flow is used to initiate the imitative game. The main idea consists in using a pattern recognition system in order to allow the student to continue its imitative game even if the teacher is stopped. Since the movement detection and the pattern recognition systems work in parallel, they can provide different answers with different time constant. Neural fields equations are used to merge these information and to allow a stable dynamical behavior of the robot. Moreover, the stability of the decision making allows the robot to online learn to recognize the teacher from one image to the next.