The paper proposes a parameter identification method for a vehicle model using real measurements of on-board sensors. The motivation of the paper is to improve the localization of the vehicle when the accuracy of the regular methods is poor, e.g. in the case of unavailable GNSS signals, no enough feature for vision, or low acceleration for IMU-based techniques. In these situations the wheel encoder based odometry may be an appropriate choice for pose estimation, however, this method suffers from parameter uncertainty and unmodelled effects. The utilized vehicle model operates with dynamic wheel radius. The proposed identification method combines the Kalman-filter and least square techniques in an iterative loop for estimating the parameters. The estimation process is verified by real test of a compact car. The results are compared with the nominal setting, in which there is no estimation.