Avisual based controlmethod based on coupling a camera with a cable-driven parallel robot (CDPR) is proposed in this paper. This work is motivated by the need of overcoming the robot uncertainties resulting from cable vibrations or inaccurate initial configurations. The poses taken by the end-effector are recorded in real time and used as feedback for the control loop to enhance the robot accuracy. Experimental tests were conducted on a planar CDPR dedicated to rehabilitation and allowing to assist the patient's upper limb along a preselected path. The reference path of the prescribed exercise was recorded, for a healthy volunteer, using a motion capture system. The implementation results show a remarkable increase in the robot accuracy. In addition, thanks to the proposed correction, the initial pose of the end-effector has no longer an impact on the robot behavior. The prescribed task can be performed independently of the starting pose.