Device-free localization (DFL) plays an increasingly dominant role in some security and military applications, aiming at locating targets without carrying any extra devices. Most research utilize received signal strength, causing poor accuracy due to insufficient information. In this letter, we propose DF-CSPG, a new DFL algorithm using compressive sensing (CS) theory and potential game (PG) approach to exploit channel station information (CSI). It is the first time that the PG approach is applied in DFL. At first, the CS-based localization model is built by exploiting CSI and then investigated from a game-theoretic perspective in which the subcarriers are assumed as players. Then, we prove that the proposed DFL game is an exact PG. Based on these, the best response theory is utilized to obtain the equilibriums. Finally, simulation results show that DF-CSPG can achieve more accurate and robust performance.