In order to improve the accuracy of the radar registration method in the process of large error primary radar measurement, taking the track data of the broadcast automatic correlation monitoring system as the true value, the characteristics of the primary radar error distribution are analyzed, and a realtime radar registration method is proposed based on the improved track tracking filter. This method is intended to improve the bias function in the variable structure interactive multi-model algorithm (EMAIMM) based on the expected increase model by adopting the iterative nearest point method (RANSACICP) of random sampling consistency in point cloud registration method, effectively reducing the influence of random errors in the registration process, obtaining the optimal bias parameters, and improving the accuracy of large error primary radar registration. The results show that the average absolute error of pitch Angle and azimuth Angle can be reduced to 0.04° and 0.07° respectively, and this provides a new method for radar system error registration.