A scheme based on high-order statistical analysis of vertical axle box acceleration (ABA) is proposed to locate and identify mortar voids in ballastless tracks quickly. The ABA signal near a void area contains abnormal non-Gaussian components, but are easily buried by random noise. A 4th order cumulative time-varying kurtosis difference (TKD-cum4) method is designed to locate void areas. Twenty-four edge-type and concealed-type mortar void cases are identified and classified by bispectral analysis of ABA signal segments. The measured and simulated data verify the proposed scheme's CAM void location and recognition effect. The results show that the maximum void area location error along the railway line is 0.137 m. The recognition accuracy of unidirectional void level is 92%. The simultaneous recognition rate of void type and damage degree is nearly 90%. The good noise resistance of the proposed detection scheme is verified.