At dynamic situation, acquiring the dynamic pose error of machine tools in real time is one of the key factors to improve the accuracy of machine tool. For multi-axis machine tool, two path lasers are used to track the space motion trajectory of two measuring points on the main-spindle measuring rod in real time. Two measuring points of 3D data are obtained and two space circles are fitted out. Then the line connecting the centre of two circles is used to identify the direction and position error of the tool. As the real-time ram errors moving along Z-axis changes with the change of the thermal error and dynamic motion error caused by the change of parameters, and the ram errors affect the final position of the tool direction. The ram errors should be predicted, measured, and corrected with the adaptive Kalman filter algorithm. The original error parameters in the error compensation model are replaced by the calculated optimal estimation value to correct the pose error of the tool. Finally, the error correction compensation and the uncompensated process are applied for the frustum specimen and the S-type workpiece. The comparison of the accuracy of the measured workpiece shows that the real-time measuring method of the tool position error and the adaptive correction compensation can effectively increase the processing accuracy of five-axis machine.