Abstract: Urban surveillance videos, owing to their wide distribution and easy accessibility, provide a new data source for rapid post-earthquake seismic motion perception. However, the seismic responses of roads or scene surfaces obtained from conventional monocular surveillance videos are not repetitive observations of the same physical quantity as the ground-floor responses recorded by strong-motion seismographs. Instead, they correspond to the dynamic responses of the same earthquake event under different observation systems. Consequently, direct point-by-point time-domain waveform reconstruction is difficult to achieve, making it challenging to directly evaluate the reliability of video-based seismic response identification results. In addition, urban surveillance videos in earthquake scenarios are often accompanied by pronounced camera ego-motion, which superimposes a global common-mode component onto the target scene's actual seismic response in the observed pixel displacements, further hindering reliable vibration extraction. To address this issue, this study proposes a monocular vision-based seismic motion spectrum identification and preliminary intensity estimation method via common-mode motion decoupling. The proposed method employs spatiotemporal joint-attention visual tracking to obtain long-duration pixel displacements, suppresses camera background motion through common-mode motion decoupling, and restores the physical scale of visual displacements via pixel-to-physical space mapping. On this basis, a hierarchical evaluation framework oriented toward cross-observation-system frequency-domain consistency is established. The framework focuses on the consistency of dominant spectral structures across observation systems rather than strict reconstruction of time-domain waveforms. Taking the 2025 Yongqing M4.2 earthquake in Hebei Province as a case study, urban surveillance videos from Haidian District, Beijing, together with strong-motion seismograph records from the same region, were selected for validation. The results show that within the stable evaluation frequency band of 0.5—2 Hz, the video-response spectra exhibit good consistency with the dominant spectral structures of strong-motion records. The spectral correlation coefficient of typical segmentsexceeds 0.9, while the absolute deviation in frequency-band energy proportion does not exceed 7.3%. Ablation experiments further demonstrate that common-mode motion decoupling primarily improves the consistency of dominant spectral structures across observation systems rather than the strict reconstruction of point-by-point time-domain waveforms. The preliminary intensity estimation results based on video responses are within ±1 intensity unit of the macroseismic intensity, validating the potential of the proposed method for rapid post-earthquake intensityassessment. By leveraging existing video surveillance networks, the proposed method enables rapid post-earthquake preliminary intensity estimation without the need for pre-installed sensors, thereby providing a low-cost and scalable complementary solution for rapid seismic response perception in complex real-world scenarios.