Electrical Impedance Tomography is a non- invasive medical imaging technique that can reconstruct the conductivity distribution of a conductive domain using a small amount of voltage data. With the rapid development of artificial intelligence, it has shown unique advantages for tactile sensors used in human-robot interaction. To improve the performance of sensors based on this method, researchers have proposed methods such as increasing the number of electrodes and introducing internal electrodes, which have achieved good results. However, the complexity of the data acquisition system and the computational cost increase at the same time. In this paper, we present a new adaptive current driving method for large-area tactile sensors. This method consists of a global scanning pattern and a local optimization pattern. The global scan pattern is used in the initialization state for fast imaging of external stimuli by only a small number of boundary electrodes. The localization is performed by mass extraction, and different local optimization patterns are adaptively selected according to the location of the external stimulus in real-time. The current density near the local stimulus location is maximized to enhance the sensor performance. To verify the effectiveness of the proposed method, we prepared a large-area tactile sensor with 24 electrodes and built a data acquisition system for validation experiments. The experimental results show that our proposed method can maximize the spatial resolution of the large-area tactile sensor without compromising the real-time performance.