In dense vehicle-perception environments, the lack of coordination among vehicular radar transmissions leads to severe mutual interference, resulting in false target generation and failure in tracking actual objects. To overcome these challenges, this paper proposes a radar anti-interference target-tracking method based on the Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter. Initially, time-frequency transformation techniques are employed to extract salient signal features, revealing how interfering signals distort genuine radar echoes. Recognizing that interference intensity causes time-varying detectability of targets, the proposed approach departs from traditional explicit state-measurement association strategies. Instead, both state and measurement sets are modeled within the Random Finite Set (RFS) framework, and an adaptive association weight is introduced to represent their implicit correlation. To further leverage the spatiotemporal distribution characteristics of true and false targets, the GM-PHD filter is utilized to perform multi-target tracking and false target suppression under dynamically changing target counts. Real-world validation is conducted using a TI millimeter-wave radar platform, where an identically tuned radar placed in front of the test vehicle introduces deliberate RF interference. Under these conditions, the radar′s noise floor is significantly elevated and the signal-to-noise ratio drops below -10 dB. Experimental results demonstrate that the proposed method maintains robust and accurate tracking performance in challenging scenarios, including target crossing and occlusion. Compared with benchmark algorithms, it achieves approximately a 50% reduction in tracking error, thereby validating its effectiveness and strong anti-interference capability in practical vehicular environments.