This paper proposes a visual servoing system based on a projection-based recognition method. The projection-based visual servoing system using only two calibrated cameras can identify the position and orientation of arbitrary-shaped unknown objects. For a visual servoing system, it needs to know the spatial dimension data of the observed object. However, in the detection of an unknown object, the visual servoing system generally directly obtains 2D images of the observed object, and doesn't know the three-dimensional size. Especially the detection of distance is important. Through an inverse projective transformation, we have realized the spatial transformation of 2D to 3D. And a genetic algorithm which is called“ Real-Time Multi-Step Genetic Algorithm(RM-GA)”is used to solve on-line optimization problems. In this paper, the tracking ability of projection-based visual servoing method has been verified through arbitrary-shaped object tracking experiments. The results of experiments will be reported.