Robust Person-Following Based on Monocular Vision for Elderly-Care Robot

Abstract

The Elderly-Care Robot (ECR) addresses the aging population issue by assisting elderly individuals with recognition, tracking, and companionship, easing the burden on healthcare systems and families. Its person-following functionality consists of visual tracking and visual servo components. Current visual tracking algorithms face challenges like occlusion and limited computational resources. This paper proposes a robust tracker for embedded systems by combining Kernelized Correlation Filter (KCF) and Kalman Filter (KF). The approach utilizes the Average Peak-to-Correlation Energy (APCE) index to detect occlusions and adapt the learning rate accordingly. For visual servoing, an image-based visual servo control algorithm is developed to estimate the robot's position and guide its movement using monocular vision. Experiments on datasets show improved occlusion handling and real-time performance with an overlapping success rate of 0.602 on OTB-100 and 0.247 on LaSOT. In addition, real-world tests demonstrate the robot's effectiveness in tracking and maintaining distance to a target in an obscured environment with a tracking speed of 0.586 m/s, while maintaining a safe distance of 2.36 m, achieving an 87% success rate. © 2025 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.