Aiming at the problem that the current security patrol robot system has a high requirement for the load capacity of the robot master controller, a distributed security patrol robot system based on robot operating system (ROS) is designed. The simultaneous Localization and mapping (SLAM) nodes and the vision recognition nodes are distributed in the robot host and the remote monitor in the local ROS network. On the SLAM side, the gmapping algorithm is used to realize the mapping in the dynamic environment and the localization of the robot; In the aspect of visual recognition, TensorFlow deep learning framework is used to recognize multiple objects in the scene by using TensorFlow Object Detection API. The results show that the system provides a reference for the design of ROS-based distributed security patrol robot system.