Real-time and accurate detection of the voltage sag characteristics is the premise to achieve dynamic voltage restorer compensation. An improved αβ-dq transformation detection method is presented for the limitations of traditional detection methods. In this method, the α-axis component of the αβ static coordinate system is deduced according to the single-phase voltage. The virtual β-axis component is constructed from the derivative of the α-axis component. The magnitudes, duration, phase-angle jump of the voltage sag are detected quickly and accurately by αβ-dq transformation and low-pass filter. The original data is real-time, which ensures faster detection response speed and reduces the computation greatly. In addition, an optimization design method for digital low-pass filter is presented against the contradictions existing in real-time and filtering effect of common low-pass filter. This adopts inertial filter to improve the characteristics of Butterworth low-pass filter and enable them to better adapt to the needs of the voltage sag detection thus improving the real-time quality and precision of dynamic voltage restorer.