3D Gaussian Splatting has emerged as a promising technique for high-quality\n3D rendering, leading to increasing interest in integrating 3DGS into realism\nSLAM systems. However, existing methods face challenges such as Gaussian\nprimitives redundancy, forgetting problem during continuous optimization, and\ndifficulty in initializing primitives in monocular case due to lack of depth\ninformation. In order to achieve efficient and photorealistic mapping, we\npropose RP-SLAM, a 3D Gaussian splatting-based vision SLAM method for monocular\nand RGB-D cameras. RP-SLAM decouples camera poses estimation from Gaussian\nprimitives optimization and consists of three key components. Firstly, we\npropose an efficient incremental mapping approach to achieve a compact and\naccurate representation of the scene through adaptive sampling and Gaussian\nprimitives filtering. Secondly, a dynamic window optimization method is\nproposed to mitigate the forgetting problem and improve map consistency.\nFinally, for the monocular case, a monocular keyframe initialization method\nbased on sparse point cloud is proposed to improve the initialization accuracy\nof Gaussian primitives, which provides a geometric basis for subsequent\noptimization. The results of numerous experiments demonstrate that RP-SLAM\nachieves state-of-the-art map rendering accuracy while ensuring real-time\nperformance and model compactness.\n