Vision-Based Real-Time 3D Position Sensor for Spherical Actuators Using ICP Pose Estimation

초록

This paper presents a vision-based real-time 3D position sensor for spherical actuators using Iterative Closest Point (ICP) pose estimation. Unlike conventional rotary motors with fixed axes, spherical actuators enable omnidirectional rotation, which makes traditional one-dimensional sensing methods, such as encoders, Hall sensors, and resolvers, unsuitable for accurate orientation measurement. To address this challenge, we propose a non-contact sensing approach that reconstructs three-dimensional surface point clouds from monocular camera images and estimates the full 3D orientation of the spherical rotor. Arbitrary and non-uniform visual patterns are applied to the rotor surface, and both intentional patterns and natural surface imperfections (e.g., scratches, dust, and wear) are exploited as geometric features for pose estimation. Edge features are extracted using Canny edge detection, reconstructed into 3D point clouds through a geometrically consistent mapping model, and aligned via ICP to estimate the spherical angles θ and ϕ. Experimental validation is conducted using a spherical rotor with a diameter of 15 cm. Under single-axis rotations, a minimum mean absolute error of 0.45° in θ is achieved after pixel-per-revolution calibration. Under dynamic multi-axis rotational motions, the estimated tangential azimuth direction ϕ exhibits compact zero-centered error distributions within approximately 0.2°– 0.9°, confirming consistent 3D directional estimation. Long-term stability analysis further shows that cumulative angular error accumulation remains approximately 0.15° per degree under appropriate sensing conditions. A computational architecture comparison demonstrates the feasibility of real-time operation. The complete processing pipeline, including image preprocessing, 3D reconstruction, and ICP-based pose estimation, achieves a total processing time below 32 ms per frame, satisfying real-time requirements at 30 fps. These results confirm that the proposed vision-based sensor provides a robust, non-contact, and practical solution for real-time 3D orientation sensing in spherical actuator systems. © 2001-2012 IEEE.

키워드