Online Marker-Free Extrinsic Camera Calibration Using Person Keypoint Detections

Calibration of multi-camera systems, i.e. determining the relative poses between cameras, is a prerequisite for many tasks in computer vision and robotics. Camera calibration typically achieved using offline methods that use checkerboard targets. These methods, however, often are cumbersome lengthy, considering new required each time any camera pose changes. In this work, we propose novel, marker-free online method extrinsic multiple smart edge sensors, relying solely on 2D human keypoint detections computed locally sensor boards from RGB images. Our assumes intrinsic parameters to be known requires priming with rough initial estimate poses. The person views received at central backend where they synchronized, filtered, assigned hypotheses. We these hypotheses repeatedly solve optimization problems form factor graphs. Given suitable observations one or persons traversing scene, estimated converge towards coherent within few minutes. evaluate our approach real-world settings show achieves lower reprojection errors compared reference generated by an traditional target.