To address the issue of low pose estimation accuracy or complete failure of visual simultaneous localization and mapping （SLAM） algorithms that relied solely on single-point features in indoor environments characterized by sparse texture and varying illumination conditions， a pose-decoupled RGBD-SLAM system based on point， line， and plane features was proposed. This system leveraged the complementary advantages of different features and the structured characteristics of the scene， employing the concept of pose decoupled estimation. By utilizing the Manhattan World hypothesis， drift-free rotation estimation was achieved， while translation was estimated through the minimization of a multi-feature joint error function. This approach mitigated the cumulative error effects associated with traditional SLAM systems that employed frame-by-frame tracking， thereby enhancing the accuracy of pose estimation and facilitating the construction of a richly informative point-line-plane structural map of the environment. Experimental results indicate that the proposed SLAM system achieves an average absolute trajectory accuracy improvement of 54.5%， 23.5%， and 28.3% compared to ORB-SLAM2， PL-SLAM， and SP-SLAM， respectively， across eight subsequences of the ICL-NUIM dataset. Additionally， in eleven subsequences of the TUM RGBD dataset， improvements of 33.9%， 26.2%， and 11.7% are observed， demonstrating superior global localization performance and enhanced system robustness. Furthermore， the constructed point-line-plane structural map provides a more comprehensive representation of the environment.