Modern advanced robotic hand prostheses face many challenges with respect to user control of the devices, which can often result in their rejection by users. A vision-based control system is proposed, which could bypass many of these issues by having the computational component of the system control the grasping of the hand based on the spacial relation of the prosthesis and the grasp object as provided from object tracking. This eliminates complicated control and the need for training for use of the prosthesis. In this thesis a control system that consists of a camera, image processing software, and other software to handle robot control is explored. The particular version realized for the study uses a webcam paired with Vuforia software in the Unity environment to track the prosthesis and the grasp object, while the SynGrasp toolbox for MATLAB is used to control a virtual hand. Successful use of the proposed technology would require: 1) accurate object recognition, 2) determining what object a user is reaching to, 3) predicting user's arm motion for the near future, 4) measuring the kinematics of the arm, hand, and object relative to each other, and 5) controlling the hand to close around the target object such that an adequate grasp is achieved. This thesis focuses on Points 4 and 5, though the other points are addressed in summary form. Therefore, in this context, feasibility entails accurate, camera-based kinematics measurement and appropriate grip selection for a set of target objects. Kinematic errors were determined under realistic static and dynamics conditions. Quantifiable systematic errors were removed through a calibration processes leaving only zero mean random errors to characterize. It was determined that random translation error typically had a standard deviation of about half a millimeter when stationary and nine millimeters when in realistic worst case motion. Random rotational error was found to have a standard deviation of about half a degree. These error estimates were then propagated through grasp selection algorithms via the SynGrasp toolbox for MATLAB. Simulations with small magnitude error observed in static or quasi-static scenarios resulted in small differences in grasp compared to the control case with 73 to 99 percent of simulations maintaining 90 percent or better relative performance quality compared to its control case. Simulations with larger magnitude error observed in dynamic scenarios resulted in significant differences in grasp compared to the control case with 10 to 30 percent of simulations maintaining 90 percent or better relative performance quality compared to its control case. The stationary simulation results were deemed to be more relevant due to most grasp situations a prosthesis could encounter better relating to a static or quasi-static environment than one in motion. Thus due to the stationary simulation results, the particular embodiment of the concept implemented in this study was ultimately ruled to be feasible.