A Two-Stage Monocular Vision Detection Method for 6D Pose Estimation in Multi-Heterogeneous Robot Systems

Multi-robot systems are crucial for enhancing work efficiency, flexibility, and adaptability, with the accurate acquisition and processing of pose information in real-world environments being a prerequisite for their implementation. This study proposes a 6D pose visual localization system for heterogeneous multi-robot systems based on monocular vision, addressing the issues of high costs and suboptimal real-time performance in existing systems. The main contributions include the design of a cost-effective and efficient monocular vision localization system framework, the proposal of an enhanced YOLOv8 detection model that integrates object classification, keypoint detection, and 6D pose analysis, and the development of a 6D pose analysis method based on 3D model-based PnP keypoint threshold screening and matching, thereby improving the system’s operational speed and accuracy. This research achievement can be applied in multiple fields such as multi-heterogeneous robot system coordination verification, autonomous navigation, and intelligent manufacturing.