Development of a Real-Time Multi-Person 3D Keypoint Detection System Using Stereoscopic Cameras and RTMPose
Taufik Soesilo, Praveen Nuwantha Gunaratne, Hiroki Tamura · Proceedings of International Conference on Artificial Life and Robotics · 2025
The feature offers a way of tracking the movement patterns of many people at once, and is critical in occupational health, sports performance, and team-based work settings.The selected traditional biomechanical analysis systems, in turn, are largely considered in detecting single person movement, which weakens their applicability to movement analysis that involves interacting with multiple people.In this paper we consider a real time multi-person detection and analysis system using stereoscopic cameras and RTMPose, a novel high real-time pose estimation framework.RTMPose offers real time analysis of 2D key points for the individuals and this data is later augmented with depth data coming from stereoscopic imaging to give 3D skeletal data.The benefit of employing RTMPose is that the system is able to perform accurate and fast multiple persons tracking despite present occlusion scenarios.Consequently, the system overcomes the drawbacks of prior methods, including reliance on wearable devices and unsuitability for out-of-door environments, by employing stereoscopic cameras and RTMPose with low-latency and high-accurate inference.Experimental results demonstrate the system's ability to provide detailed real-time analysis of posture and movement for multiple individuals in diverse scenarios.This research highlights the potential of RTMPose-powered systems to advance multi-person biomechanical analysis for applications ranging from workplace monitoring to sports performance assessment.