Study on CCTV Utilization for Autonomous Mobile Robots to Avoid Dynamic Obstacles

This study proposes a vision-based obstacle recognition system that supports the safe navigation of Autonomous Mobile Robots (AMRs) by utilizing fixed external CCTV cameras and deep learning-based object detection technologies. With the recent proliferation of smart logistics and automation environments, there is a growing need for rapid and accurate response to dynamic obstacles. The proposed system is designed to meet these demands by integrating object detection using YOLOv4, mobility assessment based on optical flow, motion prediction using a Kalman filter, and 2D map projection via homography transformation. Through this framework, the system can identify the type, position, velocity, and direction of obstacles in real time and transmit this information to the AGV system using ROS and a ZeroMQcommunication structure, thereby supporting safe and predictable path planning. Furthermore, the system is designed to satisfy the major safety requirements outlined in ISO 3691-4. Experimental results demonstrate that the system can overcome the limitations of traditional onboard sensor-based systems, particularly in handling dynamic obstacles, compensating for blind spots, and extending external perception capabilities. This study contributes to securing safety in industrial AMR applications and suggests the potential for further expansion into various industrial environments.