Simultaneous localization and mapping (SLAM) is one of the fundamental challenges for motion robot systems. Over time, SLAM methodologies have evolved from conventional filter-based approaches to contemporary optimization approaches. SLAM based on the particle filter has been widely used in mobile robot technology, especially in 2D ground robot SLAM. However, traditional implementation of particle filter visual SLAM algorithm typically only considers 2D position states, overlooking the orientation along the z-axis. Moreover, they entail substantial computational overhead as the state needs to be updated once there is a new visual observation received. To address these limitations, this paper presents a novel 2D visual SLAM framework that incorporates three freedoms and leverages Aruco Marker as the robust observation. A distance sliding window is introduced to avoid the computational workload resulting from the expansion of state dimensionality. Several simulation experimental results demonstrate that the proposed algorithm significantly enhances computational efficiency while maintaining accuracy and robustness in both localization and mapping tasks, through a reduced requirement for visual updates. The whole implementation is open source ( https://github.com/Happy-ZZX/Puzzlebot ).