Abstract: ‍ ‍Owing to the rapid development of autonomous driving technology， 4D millimeter-wave radar has become a key sensor for simultaneous localization and mapping （SLAM） owing to its all-weather adaptability and anti-interference capabilities. However， in confined environments such as tunnels， multipath effects result in false-positive point clouds that severely affect the positioning accuracy and mapping quality of radar SLAM systems. To address this issue， this paper proposes a novel sliding-window dynamic filtering algorithm based on the analysis of the characteristics of millimeter-wave radar point-cloud data and scattering-angle features in tunnels. The method combines the spatial statistical characteristics of point clouds with neighborhood-density detection techniques to remove outlier-noise point clouds. It utilizes coarse radar point-cloud registration to obtain a prior estimated pose and incorporates three-dimensional pitch and azimuth scattering-angle features to distinguish and cluster the actual target point-cloud data. Subsequently， the random sample consensus algorithm is applied to fit the tunnel-wall plane and construct the tunnel-wall model. By introducing a dynamic sliding-window update strategy， the fitted tunnel-wall model and prior estimated pose are used to update the current pose node’s distance to the wall boundary in real time. Additionally， a distance threshold is used to further eliminate false-positive point clouds and noise point clouds outside the tunnel space. Global pose correction and local map updates are completed within a factor-graph-optimization framework. This study conducts experimental verification using multiple datasets obtained in actual tunnel environments under different scenarios. The experimental results show that the proposed method effectively reduces false-positive point-cloud interference， significantly improves positioning accuracy and mapping quality， and maintains high stability in complex environments. This study provides new technical insights and implementation pathways for improving the robustness of 4D millimeter-wave radar SLAM in confined environments.