Structured illumination microscopy (SIM) exhibits high precision and robust measurement performance on complex surfaces. Continuous vertical scanning structured illumination microscopy (CVS-SIM) reduces the number of structured illumination patterns through continuous scanning to enhance measurement efficiency and has gained widespread attention. However, constrained by limited acquisition information, CVS-SIM suffers from a low signal-to-noise ratio and degraded reconstruction accuracy in complex surface topography reconstruction due to dynamic background intensity fluctuations, non-uniform surface reflectance characteristics, and substantial local slope variations. To solve the above problems, this study proposes a robust optical sectioning method integrating sliding-window background estimation with Hilbert transform-based contrast-weighted HiLo image fusion. The proposed method employs sliding-window averaging filtering and adaptive band-stop filtering to estimate dynamic background intensity while suppressing residual noise interference. Furthermore, it utilizes the Hilbert transform to decode and separate carrier and envelope signals, combined with a fringe contrast-weighted HiLo fusion strategy, effectively reducing noise impact on optical sectioning. Experimental results demonstrate that for a through-glass-via sample with steep local slopes, the proposed method achieves a higher signal-to-noise ratio (SNR) in axial modulation response compared to conventional methods, reducing reconstruction error from 1.458 µm to 0.104 µm. For the samples with complex textures and reflectivity variations, such as redistribution layer structures and interdigitated electrodes, the method maintains measurement accuracy and lateral resolution while suppressing surface measurement noise by over 51%. Comprehensive testing across various samples confirms that the proposed method achieves consistent high-quality surface reconstruction across diverse surface topographies.