The location service is foundation of pervasive computing, intelligent transportation and application of WSN, so the wireless positioning is the focus of the current research. Current location techniques include ToA, TDoA and RSSI. And considering that non-line-of-sight propagation introduces the significant bias in measurements and thus the very large error in current location estimation technique, a two-step position tracking scheme is proposed to mitigate the position tracking error. In this scheme, an SVR model of node position with radio parameters is firstly established by supervising learning. Then the position estimation from SVR model is smoothed by the game. Because the radio parameter is not considered as the distance measure, but as the feature to train the SVR model, the side effect of non-line-of-sight is mitigated. And by modeling the noise as the adversary of position estimator, a position smooth method based on game theory is introduced. Because of its ability to handle the uncertain and unmodeled noise, game theory can attain in theory more accuracy of position tracking compared with Kalman-filter, which can only deal with the Gaussian noise. Simulations show that the scheme proposed results in the more accurate performance of location estimation, especially in the harsh environment.