This paper evaluates the performance of the Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter for tracking multiple targets in an intelligent vision system. A stereo vision camera is used to get the left and right image sequences in order to extract 3-D coordinates of the targets' positions in the real world scene. The 3-D trajectories of the targets are tracked by a GM-PHD filter. Moreover, the label continuity of the targets is guaranteed by a new method of labeling. Motion speed and angular velocity are proposed for the evaluation of the accuracy and label continuity of the filter in the implemented 3-D test motion model. The simulation results for two moving targets show that the proposed system not only robustly tracks them, but also maintains the label continuity of the two targets.