Electrostatic levitation is one of the most important methods used to study the properties of materials without having to collide them against the wall of a chamber. A dynamic, ground electrostatic levitation control system model is designed and built using two Graphics Processing Units (GPUs) to process the image sequence and calculate the position of the material. In addition, a real-time detection algorithm is proposed for quick real-time visual detection of the targets. The melting experiment of the material is realized with an open-loop and Proportional-Integral-Derivative (PID) controller. This eliminates the charge supplement device of the deep ultraviolet lamp, and an image processing speed of 700 frames per second at the resolution of 304 pixels × 304 pixels and control precision of approximately ±0.02 mm are achieved. Furthermore, the relevant parameters concerning the control effects between the simulation model and experiments are consistent. For a disturbance of 900 V, which causes the acceleration of the material to reach approximately 1.274 m/s2, the system can stabilize in 340 ms. Therefore, the feasibility and reliability of an electrostatic levitation control system based on high-speed vision and the accuracy of its dynamic model were proved.