With signal and noise state model of Kalman filter, the frequency was unbiasedly estimated, which was suitable for accurate measurement of locomotive speed. Time-varying window was used to capture pulse signals to overcome shortcomings of conventional algorithm such as instable refresh time and measurement accuracy, as well as to ensure the accurate measurement of different frequencies. FPGA-based algorithm design and simulation has demonstrated that with Kalman filter and time-varying window, it was effective and accurate to measure locomotive rotate speed.