With the development of autonomous driving technology, automotive radar has received unprecedented attention due to its day-and-night and all-weather working capability. It is worthwhile to note that more and more vehicles are equipped with automotive radars, resulting in mutual interference between radars. Interference degrades radar target detection performance and makes sensed information unreliable, so interference suppression is necessary. In this paper, a novel method based on power-weighted Hough transform is proposed for mitigating the radar mutual interference and improving the safety of autonomous driving systems. Firstly, the frequency modulation characteristics of interference signals and target echo signals are analyzed, and differences between the two signals are presented. Secondly, based on the straight line detection technique, the interference power in the time-frequency domain is accumulated, and the interference is accurately detected and localized. Finally, the target echo is recovered by an autoregressive model. Compared with existing state-of-the-art methods, the proposed method has the ability to retain more useful signals after interference mitigation, and achieve better interference detection robustness under low signal-to-noise ratio conditions. Simulation experiments and real scenario experiments verify the effectiveness of the proposed method and show its superiority.