To address the need for real-time solutions when handling kinematic baselines using multi-system and multi-frequency data, where both the rover and reference stations are in motion, we propose a novel non-combined double-difference approach based on extended Kalman filtering. By employing single-difference ambiguity conversion, this approach achieves ambiguity resolution and avoids the frequent alteration of reference satellites in kinematic baselines. Analysis of actual measurement data demonstrates that within a 20 km baseline length range, this algorithm achieves real-time horizontal and vertical baseline accuracies of approximately 1 cm and 2 cm, respectively. The success rate of ambiguity resolution nears 100%, surpassing conventional ionospheric-free double-difference models. Furthermore, as more systems and frequencies are introduced, this proposed method consistently provides superior real-time solutions for kinematic baselines. The practical effectiveness of this approach is further validated through real-flight experiments involving two unmanned aerial vehicles.