Indoor mobile robots have a wide range of application value, such as home cleaning, indoor exploration, etc. SLAM technology is usually used to solve its localization and mapping problems, with VSLAM (visual SLAM) being a key research focus. Traditional VSLAM robots mainly rely on processors on the vehicle body. This architecture of pure on-board processor leads to problems such as low battery endurance and high vehicle body cost. In addition, traditional VSLAM frameworks, such as ORB-SLAM3, tend to exhibit degraded performance in environments with dynamic objects. This research presents a novel approach to address these challenges by designing an indoor distributed VSLAM system that offloads computationally intensive tasks to an off-board station via advanced wireless communication technology (WIFI 6) and hardware-accelerated video encoding (H.264). Additionally, the system integrates deep learning-based image segmentation to filter dynamic objects, thereby improving tracking error. The research employs ORB-SLAM3 as the foundational VSLAM framework, combined with ROS 2 for distributed communication management. The implemented system demonstrates real-time communication performance with a Sense-Move delay of 64.9 ms, network delay of 19.1 ms, and network throughput of 0.5 MBps. The integration of YOLO-based semantic segmentation for dynamic object filtering substantially improves trajectory tracking accuracy, reducing RMSE from 0.669 m to 0.015 m in dataset evaluations. The integrated system achieves 10–15 fps on Jetson Orin Nano Super hardware. This distributed architecture successfully reduces the computational burden on indoor mobile robots while maintaining real-time operation, demonstrating the effectiveness of integrating advanced wireless technologies, hardware-accelerated video encoding, and deep learning.